CN102142128A - Method and device for computing description vectors of interest points of image - Google Patents

Abstract

The embodiment of the invention provides a method and device for computing the description vectors of the interest points of an image. The method provided by the invention mainly comprises the following steps: selecting a window region in the Gaussian scale space in which the interest points are positioned, wherein the window region is divided into a set number of sub-regions; selecting the set number of sampling points in each sub-region; computing the position coordinates and derivative coordinates of each sampling point; partitioning each sub-region into the set number of branch regions; computing and obtaining the description vector factors corresponding to each sub-region according to the computing method corresponding to the partitioning method of the set number of the branch regions and the position coordinates and derivative coordinates of each sampling point; and integrating the description vector factors corresponding to all the sub-regions and obtaining the description vectors of the interest points. The invention can obviously reduce the computation amount of the description vectors of the interest points of the image.

Description

The computing method and the device of the description vector of the point of interest of image

Technical field

The present invention relates to communication technical field, relate in particular to the computing method and the device of description vector of the point of interest of a kind of image in the Flame Image Process.

Background technology

Images match is to analyze and find out corresponding zone in twos in the multiple image that Same Scene is obtained, and does not need to recover the space geometry relation of image.Image matching technology is the popular coupling that just is based on local feature at present, reason is generally all to exist on how much between the image to be matched and the variation on the luminosity, all be nonlinearities change in essence, so can only find at utmost similar image sheet at subrange, by these image sheets are described, realize the coupling of image local.Image matching technology based on local feature mainly comprises: the generation of the description vector of the detection of the point of interest of image and point of interest, above-mentioned point of interest are the pixel that has notable feature in the area-of-interest of image.

The generation method of the description vector of a kind of point of interest of the prior art is: SIFT (ScaleInvariant Features Transformation is based on local feature description's method of gradient vector spatial statistics distribution) method.The processing procedure of this method mainly comprises:

1, select a zone near the point of interest position, the size in this zone is the linear function of point of interest characteristic dimension;

2, above-mentioned zone is divided into 4 * 4 totally 16 sub regions (sub-segion);

3, in each subregion, choose a sampled point, calculate the gradient magnitude and the direction of each sampled point, and the gradient magnitude of sampled point is weighted, the gradient magnitude of the gradient magnitude after the weighting as sampled point;

The gradient magnitude of sampled point and direction calculating formula are as follows:

$m(x,y)=\sqrt{{(I(x+1,y)-I(x-1,y))}^2+{(I(x,y+1)-I(x,y-1))}^2}$

θ(x，y)＝tan ^-1(I(x，y+1)-I(x，y-1)/I(x+1，y)-I(x-1，y))

4, calculate the direction histogram of the subregion at each sampled point place by interpolation method, each subregion need carry out interpolation arithmetic 8 times.

According to the gradient magnitude of above-mentioned each sampled point that obtains and the direction histogram of direction and each subregion, calculate the description vector factor of each sampled point.Then, comprehensively the description of all sampled points vector factor obtains the description vector of a point of interest.

The shortcoming of the generation method of the description vector of above-mentioned point of interest of the prior art is: in the SIFT method, the size and Orientation that each sampled point all will carry out a subgradient calculates, carry out interpolation arithmetic 8 times, each interpolation will comprise 4 relevant floating-point operations with other of floating-point multiplication.And a local feature relates to the sampling number of calculating on average about 2000 to 4000.And the image of a pair 500 * 500, the quantity of its point of interest generally also is more than 1000.Therefore, in this SIFT method, the calculated amount of the description vector of calculating point of interest is bigger, and the complexity of calculating is bigger, and the formation speed of the description vector of point of interest is slow, and this SIFT method efficiency ratio is lower.

Summary of the invention

Embodiments of the invention provide a kind of computing method and device of description vector of point of interest of image, with the complexity of the calculating of the description vector of the point of interest that reduces image, improve the formation speed of the description vector that generates point of interest.

A kind of computing method of description vector of point of interest of image comprise:

In Gauss's metric space at point of interest place, choose form region, described form region is divided into sets the quantity sub regions, in each subregion, choose and set the individual sampled point of quantity, calculate the position coordinates and the derivative coordinate of each sampled point;

Described each subregion is divided into a setting quantity subregion, according to the position coordinates and the derivative coordinate of a described setting quantity pairing computing method of subregional division methods and described each sampled point, calculate the description vector factor of each subregion correspondence;

The description vector factor of all subregion correspondences is carried out comprehensively obtaining the description vector of described point of interest.

A kind of calculation element of description vector of point of interest of image comprises:

The coordinate Calculation module; Be used for choosing form region, described form region be divided into set the quantity sub regions, in each subregion, choose and set the individual sampled point of quantity, calculate the position coordinates and the derivative coordinate of each sampled point at Gauss's metric space at point of interest place;

Vectorial factor computing module is described, be used for described each subregion is divided into a setting quantity subregion, according to the position coordinates and the derivative coordinate of a described setting quantity pairing computing method of subregional division methods and described each sampled point, calculate the description vector factor of each subregion correspondence;

Describe the vector calculation module, be used for the description vector factor of all subregion correspondences is carried out comprehensively obtaining the description vector of described point of interest.

The technical scheme that provides by the embodiment of the invention described above as can be seen, the embodiment of the invention is divided into by first subregion with the sampled point place and sets a quantity subregion, calculates the vectorial factor of description of each sampled point correspondence.Thereby realize obviously having reduced the calculated amount of description vector of the point of interest of image, reduce the complexity of calculating of description vector of the point of interest of image, improve the formation speed of the description vector that generates point of interest.

Description of drawings

In order to be illustrated more clearly in the technical scheme of the embodiment of the invention, the accompanying drawing of required use is done to introduce simply in will describing embodiment below, apparently, accompanying drawing in describing below only is some embodiments of the present invention, for those of ordinary skills, under the prerequisite of not paying creative work, can also obtain other accompanying drawing according to these accompanying drawings.

The processing flow chart of the computing method of the description vector of the point of interest of a kind of image that Fig. 1 provides for the embodiment of the invention one;

The acquisition process synoptic diagram of a kind of metric space image that Fig. 2 provides for the embodiment of the invention one;

The processing procedure synoptic diagram of a kind of DOG detection method that Fig. 3 provides for the embodiment of the invention one;

The computation process synoptic diagram of the principal direction of a kind of first point of interest that Fig. 4 provides for the embodiment of the invention one;

The form region of choosing around a kind of first point of interest that Fig. 5 provides for the embodiment of the invention one and carry out shift transformation after the contrast synoptic diagram of form region;

A kind of subregion with first sampled point place that Fig. 6 provides for the embodiment of the invention one is divided into the synoptic diagram in 6 zones;

A kind of subregion with first sampled point place that Fig. 7 provides for the embodiment of the invention one is divided into the synoptic diagram in 4 zones;

A kind of subregion with first sampled point place that Fig. 8 provides for the embodiment of the invention one is divided into the synoptic diagram in 8 zones;

The structural representation of the calculation element of the description vector of the point of interest of a kind of image that Fig. 9 provides for the embodiment of the invention two.

Embodiment

For the purpose, technical scheme and the advantage that make the embodiment of the invention clearer, below in conjunction with the accompanying drawing in the embodiment of the invention, technical scheme in the embodiment of the invention is clearly and completely described, obviously, described embodiment is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills belong to the scope of protection of the invention not making the every other embodiment that is obtained under the creative work prerequisite.

For ease of understanding, be that example is further explained explanation below in conjunction with accompanying drawing with several specific embodiments, and each embodiment does not constitute the qualification to the embodiment of the invention to the embodiment of the invention.

Embodiment one

The treatment scheme of the computing method of the description vector of the point of interest of a kind of image that this embodiment provides comprises following treatment step as shown in Figure 1:

Step 11, carry out convolution algorithm with Gaussian function and original image and obtain the metric space image.

Import pending original image, carry out convolution algorithm, obtain the metric space image of a sequence with not homoscedastic Gaussian function of a sequence and above-mentioned original image.

The acquisition process of a kind of above-mentioned metric space image that this embodiment provides as shown in Figure 2.In Fig. 2, I is an original image, at first uses Gaussian function G (δ 1) (δ 1=1.6) and I to carry out convolution, obtains the metric space image I ₁, I ₁=I**G (δ 1);

Then, use Gaussian function Carry out convolution with I, obtain the metric space image I ₂, I ₂=I**G (δ 2).

Above-mentioned process of convolution process is equivalent to original color image is carried out smoothing processing.

Step 12, employing DOG (Difference of Gaussian, difference of Gaussian yardstick) detection method obtain unique point and the point of interest on Gauss's metric space.

According to above-mentioned metric space image and original image, obtain point of interest on the metric space image with the DOG detection method.The processing procedure synoptic diagram of a kind of above-mentioned DOG detection method that this embodiment provides as shown in Figure 3, this processing procedure mainly comprises:

With original image I and metric space image I ₁Carry out the difference of Gaussian computing by the DOG detection method, obtain comprising first Gauss's metric space of 9 unique points; Use the metric space image I ₁With the metric space image I ₂Carry out the difference of Gaussian computing by the DOG detection method, obtain comprising second Gauss's metric space of 9 unique points; Use the metric space image I ₂With the metric space image I ₃Carry out the difference of Gaussian computing by the DOG detection method, obtain comprising the 3rd Gauss's metric space of 9 unique points.

Pixel value to 27 unique points in above-mentioned first Gauss's metric space, second Gauss's metric space and the 3rd Gauss's metric space compares, and getting wherein the maximum or minimum unique point of pixel value is point of interest.As shown in Figure 3, the black pixel point in second Gauss's metric space is a point of interest, the characteristic dimension of this point of interest

According to above-mentioned processing procedure shown in Figure 3, and the metric space image of some other above-mentioned original image, point of interest and characteristic of correspondence yardstick on some other Gauss's metric space can be obtained.

Point of interest on the above-mentioned Gauss's metric space that obtains is the point of interest of above-mentioned original image just.

Step 13, calculate the principal direction of first point of interest.

In all points of interest, choose first point of interest.On Gauss's metric space at this first point of interest place according to the characteristic dimension δ of first point of interest, near the border circular areas that to choose a radius first point of interest be 4 δ, ask for respectively in this border circular areas the horizontal direction derivative and, the vertical direction derivative and, with above-mentioned horizontal direction derivative and, vertical direction derivative and be that the direction of the gradient of coordinate formation is exactly the principal direction of above-mentioned first point of interest.

The computation process of the principal direction of a kind of above-mentioned first point of interest that this embodiment provides as shown in Figure 4, in Fig. 4, ∑ dx be the horizontal direction derivative and, ∑ dy be the vertical direction derivative and, theta0 is the angle of the principal direction correspondence of above-mentioned first point of interest.

Step 14, choose form region around first point of interest, choose the sampled point of setting quantity in each subregion in above-mentioned form region, calculate the horizontal direction derivative I of each sampled point _{Gx '}With vertical direction derivative I _{Gy '}

In Gauss's metric space at first point of interest place, choose the foursquare form region of certain forms width around first point of interest, the characteristic dimension of above-mentioned forms width and point of interest is relevant, such as, above-mentioned forms width can for

N ∈ [8,12].

Above-mentioned forms width can not be excessive, and the forms width is crossed conference and introduced the excessive data of distortion, thereby the design original intention-part of having run counter to local feature is approximate.Above-mentioned on the other hand forms width again can not be too little, otherwise it can not provide enough information to portray this regional area.

Above-mentioned form region is divided into 1-16 sub regions (Sub-region), each subregion big or small identical.In each subregion, evenly select to set the individual sampled point of quantity, such as, 1000 sampled points selected.Choose first sampled point in first subregion in above-mentioned form region, calculate the horizontal direction derivative I of this first sampled point _{Gx '}With vertical direction derivative I _{Gy '}, and further calculate horizontal direction derivative I with first sampled point _{Gx '}With vertical direction derivative I _{Gy '}The mould value of the gradient that constitutes for coordinate

Above-mentioned first sampled point (x ', y ') corresponding

Computation process as follows:

I _{Gx '}=I _{X '+1, y '}-I _{X '-1, y '}I _{Gy '}=I _{X ', y '+1}-I _{X ', y '-1}, I _{X '+1, y '}The grey scale pixel value that expression (x '+1, y ') is located, the rest may be inferred by analogy for it.

${\left|\right|\&dtri;(x^{\&prime;},y^{\&prime;})\left|\right|}_2=\sqrt{{{\mathrm{<figure-callout\; id="2"\; label="I"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">I</figure-callout>}}^2}_{{\mathrm{Gx}}^{\&prime;}}+{{\mathrm{<figure-callout\; id="2"\; label="I"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">I</figure-callout>}}^2}_{{\mathrm{Gy}}^{\&prime;}}}={\left|\right|\&dtri;(x,y)\left|\right|}_2$

Step 15, above-mentioned form region is carried out shift transformation, calculate the position coordinates and the derivative coordinate of above-mentioned each sampled point in the form region under new coordinate system according to the principal direction of above-mentioned first point of interest.

According to the angle theta0 of the principal direction correspondence of above-mentioned first point of interest, the form region of choosing around above-mentioned first point of interest is carried out shift transformation, obtain the form region under the new coordinate system.

Calculate above-mentioned first sampled point in the form region under above-mentioned new coordinate system position coordinates (x, y) and derivative coordinate (I _Gx, I _Gy).

The computing formula of position coordinates is:

x＝cos(theta0)*x′+sin(theta0)*y′

y＝-sin(theta0)*x′+cos(theta0)*y′

Derivative Coordinate Calculation formula is:

I _Gx＝cos(theta0)*I _Gx′+sin(theta0)*I _Gy′

I _Gy＝-sin(theta0)*I _Gx′+cos(theta0)*I _Gy′

(x y) carries out normalized, that is: to the position coordinates of above-mentioned first sampled point

x ₀＝x/3δ，

y ₀＝y/3δ

If x ₀∈ [2,2] and y ₀∈ [2,2] satisfies simultaneously, so (x y) is considered as effective sampled point, otherwise, be invalid sampled point, do not do subsequent treatment.

Above-mentioned boundary value [2,2] can be selected other values in theory.If but the boundary value of selecting is the decimal type, then is inconvenient to calculate; If the boundary value of selecting is too small, then the overlay area of unique point is too small, and the proper vector of Ji Suaning is unfavorable for representing this point or description that should the zone at last; Zone radius is

If make the big as far as possible and boundary value in overlay area after the conversion round convenient calculating, the optimal boundary value is [2,2].

The form region of choosing around a kind of above-mentioned first point of interest that this embodiment provides and carry out shift transformation after form region the contrast synoptic diagram as shown in Figure 5, in Fig. 5, left side figure be the form region that above-mentioned point of interest is chosen on every side.Middle graph comprises 1-16 sub regions (Sub-region) for through the form region after the displacement coordinate conversion, and wherein the sampled point that is comprised in that embedded square all is effective sampling point.The right figure be pass through after the displacement coordinate conversion and carry out the coordinate normalized after form region.

Such as, if the coordinate x of first sampled point after the normalization ₀∈ [2 ,-1], y ₀∈ [1,2] by consulting the middle graph among above-mentioned Fig. 5, can obtain this first sampled point and belong to the 13rd sub regions.

Step 16, first subregion at above-mentioned first sampled point place is divided into 6 zones, 4 zone or 8 zones, calculates the description vector factor of first subregion correspondence.The description vector factor of all subregion correspondences is carried out comprehensively obtaining the description vector of above-mentioned first point of interest correspondence.

According to above-mentioned first sampled point through after the displacement coordinate conversion and the position coordinates in the form region after carrying out the coordinate normalized (x, y) and derivative coordinate (I _Gx, I _Gy), and the subregion at above-mentioned first sampled point place calculates the description vector factor of above-mentioned first sampled point correspondence.

In order to overcome prior art the division methods of Gauss's metric space is caused describing the problem of unstable of vector, this embodiment has proposed the splitting scheme of the subregion of new sampled point correspondence, and this new splitting scheme has taken into full account I _Gx=0 (promptly

) corresponding zone is the zone in the big mould value vector set, enlarged should the zone area, make big mould value vector all drop in the same alpha region, thereby avoided interpolation arithmetic, improved the stability of describing vector.

The embodiment of the invention is divided into 6 zones, 4 zone or 8 zones with the subregion at above-mentioned first sampled point place, the computing method of the description vector factor that different division methods is corresponding different respectively.

The synoptic diagram that first subregion at above-mentioned first sampled point place is divided into 6 zones as shown in Figure 6, according to the horizontal direction derivative and the vertical direction derivative of each sampled point in first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate.Then, the Rule of judgment corresponding respectively according to 6 subregions, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 6 corresponding respectively mould values of described 6 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 6 mould values ₁..., Mode6 ₁

The concrete computing method of the description vector factor of above-mentioned first subregion correspondence are as follows:

If I _Gx＞0

I _Gy＞λ ₁I _Gx，λ ₁∈(0，+∞)， $\mathrm{<figure-callout\; id="1"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}1=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gy＜-λ ₁I _Gx，λ ₁∈(0，+∞)， $\mathrm{<figure-callout\; id="2"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}2=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

else $\mathrm{Model}3=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

Wherein:

w _{X, y}=exp (x ²+ y ²)/(2*2 ²) be gaussian weighing function;

Be horizontal direction derivative I with sampled point _GxWith vertical direction derivative I _Gy2 norms of the mould value of the gradient that constitutes for coordinate;

Be horizontal direction derivative I with sampled point _GxWith vertical direction derivative I _GyThe spatial weighting accumulation of the mould value of the gradient that constitutes for coordinate.

Above-mentioned 1 〉=λ ₁〉=1/2, among this embodiment, λ ₁=1, purpose is exactly all to be divided into one mutually in the interval with pointing to approaching gradient with principal direction.

If I _Gx＜0

I _Gy＞-λ ₁I _Gx，λ ₁∈(0，+∞)， $\mathrm{Model}4=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gy＜λ ₁I _Gx，λ ₁∈(0，+∞)， $\mathrm{<figure-callout\; id="5"\; label="Model"\; filenames="HSA00000311734200022.png"\; state="\{\{state\}\}">Model</figure-callout>}5=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

else $\mathrm{<figure-callout\; id="6"\; label="Model"\; filenames="HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}6=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

At above-mentioned first sampled point, according to above-mentioned derivative coordinate (I _Gx, I _Gy) and above-mentioned Model1 ₁, Model2 ₁, Model3 ₁, Model4 ₁, Model5 ₁And Model6 ₁Corresponding Rule of judgment calculates above-mentioned first sampled point correspondence

Then, in above-mentioned first subregion, choose second sampled point, according to above-mentioned first sampled point correspondence

Computation process, calculate second sampled point correspondence

And the like, calculate all sampled point correspondences in above-mentioned first subregion

Then, the sampled point correspondence of the Rule of judgment of above-mentioned Model1 correspondence will be satisfied

Add up and obtain above-mentioned Model1 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model2 correspondence will be satisfied

Add up and obtain above-mentioned Model2 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model3 correspondence will be satisfied

Add up and obtain above-mentioned Model3 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model4 correspondence will be satisfied

Add up and obtain above-mentioned Model4 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model5 correspondence will be satisfied Add up and obtain above-mentioned Model5 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model6 correspondence will be satisfied Add up and obtain above-mentioned Model6 value.The description vector factor of above-mentioned first subregion correspondence is Mode1 ₁..., Mode6 ₁

Return step 14, choose successively in second sub regions in the form region of around point of interest, choosing and set the individual sampled point of quantity, according to the computation process of the description of above-mentioned first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode6 ₂

Then, 16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 6 * 16=96 ties up ₁..., Mode6 ₁, Mode1 ₂..., Mode6 ₂..., Mode1 ₁₆..., Mode6 ₁₆) ₉₆Above-mentioned

This describes vector v is the pairing description vector of above-mentioned first point of interest.

The synoptic diagram that first subregion at above-mentioned first sampled point place is divided into 4 zones as shown in Figure 7, according to the horizontal direction derivative and the vertical direction derivative of each sampled point in first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate.Then, the Rule of judgment corresponding respectively according to 4 subregions, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 4 corresponding respectively mould values of described 4 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 4 mould values ₁..., Mode4 ₁

The concrete computing method of the description vector factor of described first subregion correspondence are as follows:

1 〉=λ among Fig. 7 ₂〉=1/2, in this embodiment.λ ₂＝1。The computing method of the description vector factor of above-mentioned first subregion correspondence are as follows:

If I _Gy＞0

|I _Gy|＞|λ ₂I _Gx|λ ₂∈(0，+∞)， $\mathrm{<figure-callout\; id="1"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}1=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gx＜0， $\mathrm{<figure-callout\; id="2"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}2=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

else $\mathrm{Model}4=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

If I _Gy＜0

|I _Gy|＞|λ ₂I _Gx|λ ₂∈(0，+∞)， $\mathrm{Model}3=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gx＜0， $\mathrm{<figure-callout\; id="2"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}2=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

else $\mathrm{Model}4=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

At above-mentioned first sampled point, calculate above-mentioned first sampled point correspondence

Then, in above-mentioned first subregion, choose second sampled point, according to above-mentioned first sampled point correspondence

Computation process, calculate second sampled point correspondence

And the like, calculate all sampled point correspondences in above-mentioned first subregion

Then, the sampled point correspondence of the Rule of judgment of above-mentioned Model1 correspondence will be satisfied

Add up and obtain above-mentioned Model1 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model2 correspondence will be satisfied

Add up and obtain above-mentioned Model2 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model3 correspondence will be satisfied

Add up and obtain above-mentioned Model3 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model4 correspondence will be satisfied

Add up and obtain above-mentioned Model4 value.The description vector factor of above-mentioned first subregion correspondence is Mode1 ₁..., Mode4 ₁

Return step 14, choose successively in second sub regions in the form region of around point of interest, choosing and set the individual sampled point of quantity, according to the computation process of the description of above-mentioned first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode4 ₂

Then, 16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 4 * 16=64 ties up ₁..., Mode4 ₁, Mode1 ₂..., Mode4 ₂..., Mode1 ₁₆..., Mode4 ₁₆) ₆₄

This describes vector v is the pairing description vector of above-mentioned first point of interest.

The synoptic diagram that the subregion at above-mentioned certain sampled point place is divided into 8 zones as shown in Figure 8, according to the horizontal direction derivative and the vertical direction derivative of each sampled point in first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate.Then, the Rule of judgment corresponding respectively according to 8 subregions, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 8 corresponding respectively mould values of described 8 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 8 mould values ₁..., Mode8 ₁

The concrete computing method of the description vector factor of described first subregion correspondence are as follows:

If I _Gx＞0

I _Gy＞I _Gx， $\mathrm{<figure-callout\; id="1"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}1=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

0＜I _Gy＜I _Gx， $\mathrm{<figure-callout\; id="2"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}2=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

-I _Gx＜I _Gy＜0， $\mathrm{Model}3=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gy＜-I _Gx， $\mathrm{Model}4=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

If I _Gx＜0

I _Gy＞-I _Gx， $\mathrm{<figure-callout\; id="5"\; label="Model"\; filenames="HSA00000311734200022.png"\; state="\{\{state\}\}">Model</figure-callout>}5=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

0＜I _Gy＜-I _Gx， $\mathrm{<figure-callout\; id="6"\; label="Model"\; filenames="HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}6=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gx＜I _Gy＜0， $\mathrm{<figure-callout\; id="7"\; label="Model"\; filenames="HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}7=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gy＜I _Gx， $\mathrm{<figure-callout\; id="8"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}8=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

At above-mentioned first sampled point, calculate above-mentioned first sampled point correspondence

Then, in above-mentioned first subregion, choose second sampled point, according to above-mentioned first sampled point correspondence

Computation process, calculate second sampled point correspondence

And the like, calculate all sampled point correspondences in above-mentioned first subregion

Then, the sampled point correspondence of the Rule of judgment of above-mentioned Model1 correspondence will be satisfied

Add up and obtain above-mentioned Model1 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model2 correspondence will be satisfied

Add up and obtain above-mentioned Model2 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model3 correspondence will be satisfied

Add up and obtain above-mentioned Model3 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model4 correspondence will be satisfied

Add up and obtain above-mentioned Model4 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model5 correspondence will be satisfied

Add up and obtain above-mentioned Model5 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model6 correspondence will be satisfied

Add up and obtain above-mentioned Model6 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model7 correspondence will be satisfied

Add up and obtain above-mentioned Model7 value; The sampled point correspondence of the Rule of judgment of above-mentioned Model8 correspondence will be satisfied

Add up and obtain above-mentioned Model8 value.The description vector factor of above-mentioned first subregion correspondence is Mode1 ₁..., Mode8 ₁

Return step 14, choose successively in second sub regions in the form region of around point of interest, choosing and set the individual sampled point of quantity, according to the computation process of the description of above-mentioned first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode8 ₂

Then, 16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 8 * 16=128 ties up ₁..., Mode8 ₁, Mode1 ₂..., Mode8 ₂..., Mode1 ₁₆..., Mode8 ₁₆) ₁₂₈

This describes vector v is the pairing description vector of above-mentioned first point of interest.

Step 17, the description vector of above-mentioned first point of interest is carried out non-linear normalizing, obtain final description vector.

v ^TIt is the transposition of v.

If v ' (n)＞t then v ' (n)=t, t is non-linear inhibition threshold value, the value of t can be 0.4.

V " is exactly the final description vector of above-mentioned first point of interest correspondence after the normalization.

Return step 13, in all points of interest, choose second point of interest,, calculate the description vector of second point of interest correspondence according to the computation process of the description vector of described first point of interest correspondence of above-mentioned steps 14-step 17.The rest may be inferred, can calculate the description vector of all point of interest correspondences.

The subregion that the embodiment of the invention more lays particular emphasis on above-mentioned sampled point place is divided into 4 zone and 6 dividing region schemes, because can reduce the dimension of describing vector like this.The description vector of test shows 4 partition scheme correspondences has had good differentiation, is fit to very much to have low calculated amount to require but the performance of describing vector is had the application scenario of higher expectation.

This embodiment is divided into by first subregion with the sampled point place and sets a quantity subregion, calculates the description vector factor of each sampled point correspondence.Thereby realize obviously having reduced the calculated amount of description vector of the point of interest of image, reduce the complexity of calculating of description vector of the point of interest of image, improve the formation speed of the description vector that generates point of interest.

Embodiment two

The embodiment of the invention also provides a kind of calculation element of description vector of point of interest of image, its concrete outcome as shown in Figure 9, this device mainly comprises:

Coordinate Calculation module 91; Be used for choosing form region, described form region be divided into set the quantity sub regions, in each subregion, choose and set the individual sampled point of quantity, calculate the position coordinates and the derivative coordinate of each sampled point at Gauss's metric space at point of interest place;

Vectorial factor computing module 92 is described, be used for described each subregion is divided into a setting quantity subregion, according to the position coordinates and the derivative coordinate of a described setting quantity pairing computing method of subregional division methods and described each sampled point, calculate the description vector factor of each subregion correspondence;

Describe vector calculation module 93, be used for the description vector factor of all subregion correspondences is carried out comprehensively obtaining the description vector of described point of interest.

Described coordinate Calculation module 91 specifically comprises:

Divide processing module 911, be used to import pending original image, carry out convolution algorithm, obtain a plurality of metric space images with a plurality of not homoscedastic Gaussian functions and described original image; According to described metric space image and original image, employing difference of Gaussian yardstick DOG detection method obtains point of interest and the characteristic of correspondence yardstick on Gauss's metric space;

Choose foursquare form region in Gauss's metric space at described point of interest place, the width of described form region is provided with according to the characteristic dimension of point of interest, and described form region is divided into 16 subregions that size is identical;

Computing module 912 is used for choosing the sampled point of setting quantity in each subregion of the form region at first point of interest place, calculate position coordinates (x ', y '), the horizontal direction derivative I of each sampled point _{Gx '}With vertical direction derivative I _{Gy '}, according to the I of described first point of interest _{Gx '}, I _{Gy '}Calculate the principal direction of described first point of interest;

Angle theta0 according to the principal direction correspondence of described first point of interest, described form region is carried out shift transformation, obtain the form region under the new coordinate system, calculate described each sampled point in the form region under described new coordinate system position coordinates (x, y) and derivative coordinate (I _Gx, I _Gy).

The computing formula of position coordinates is:

x＝cos(theta0)*x′+sin(theta0)*y′

y＝-sin(theta0)*x′+cos(theta0)*y′

Derivative Coordinate Calculation formula is:

I _Gx＝cos(theta0)*I _Gx′+sin(theta0)* _IGy′

I _Gy＝-sin(theta0)*I _Gx′+cos(theta0)* _IGy′

The vectorial factor computing module 92 of described description comprises: 6 subregion computing modules 921, be used for first subregion at described first sampled point place is divided into 6 subregions, and the position coordinates of described first sampled point is that (x, y), the derivative coordinate is (I _Gx, I _Gy).According to the horizontal direction derivative and the vertical direction derivative of each sampled point in first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate.Then, the Rule of judgment corresponding respectively according to 6 subregions, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 6 corresponding respectively mould values of described 6 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 6 mould values ₁..., Mode6 ₁

The concrete computing method of the description vector factor of described first subregion correspondence are as follows:

If I _Gx＞0

I _Gy＞λ ₁I _Gx，λ ₁∈(0，+∞)， $\mathrm{<figure-callout\; id="1"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}1=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gy＜-λ ₁I _Gx，λ ₁∈(0，+∞)， $\mathrm{<figure-callout\; id="2"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}2=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

else $\mathrm{Model}3=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

Wherein:

w _{X, y}=exp (x ²+ y ²)/(2*2 ²) be gaussian weighing function;

Be horizontal direction derivative I with sampled point _GxWith vertical direction derivative I _Gy2 norms of the mould value of the gradient that constitutes for coordinate;

Be horizontal direction derivative I with sampled point _GxWith vertical direction derivative I _GyThe spatial weighting accumulation of the mould value of the gradient that constitutes for coordinate;

Described 1 〉=λ ₁〉=1/2;

If I _Gx＜0

I _Gy＞-λ ₁I _Gx，λ ₁∈(0，+∞)， $\mathrm{Model}4=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gy＜λ ₁I _Gx，λ ₁∈(0，+∞)， $\mathrm{<figure-callout\; id="5"\; label="Model"\; filenames="HSA00000311734200022.png"\; state="\{\{state\}\}">Model</figure-callout>}5=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

else $\mathrm{<figure-callout\; id="6"\; label="Model"\; filenames="HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}6=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

At described first sampled point, according to described derivative coordinate (I _Gx, I _Gy) and described Model1 ₁, Model2 ₁, Model3 ₁, Model4 ₁, Model5 ₁And Model6 ₁Corresponding Rule of judgment calculates described first sampled point correspondence

Then, in described first subregion, choose second sampled point, according to described first sampled point correspondence

Computation process, calculate second sampled point correspondence

And the like, calculate all sampled point correspondences in described first subregion

Then, the sampled point correspondence of the Rule of judgment of described Model1 correspondence will be satisfied

Add up and obtain described Model1 value; The sampled point correspondence of the Rule of judgment of described Model2 correspondence will be satisfied

Add up and obtain described Model2 value; The sampled point correspondence of the Rule of judgment of described Model3 correspondence will be satisfied

Add up and obtain described Model3 value; The sampled point correspondence of the Rule of judgment of described Model4 correspondence will be satisfied

Add up and obtain described Model4 value; The sampled point correspondence of the Rule of judgment of described Model5 correspondence will be satisfied

Add up and obtain described Model5 value; The sampled point correspondence of the Rule of judgment of described Model6 correspondence will be satisfied

Add up and obtain described Model6 value; The description vector factor of described first subregion correspondence is Mode1 ₁..., Mode6 ₁

Described description vector calculation module 93 comprises, first computing module 931, be used in second sub regions of the form region that point of interest is chosen, choosing successively the individual sampled point of setting quantity on every side, according to the computation process of the description of described first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode6 ₂

16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 6 * 16=96 ties up ₁..., Mode6 ₁, Mode1 ₂..., Mode6 ₂..., Mode1 ₁₆..., Mode6 ₁₆) ₉₆, described description vector v is the pairing description vector of described first point of interest.

The vectorial factor computing module 92 of described description comprises: 4 subregion computing modules 922, be used for first subregion at described first sampled point place is divided into 4 subregions, and the position coordinates of described first sampled point is that (x, y), the derivative coordinate is (I _Gx, I _Gy).According to the horizontal direction derivative and the vertical direction derivative of each sampled point in first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate.Then, the Rule of judgment corresponding respectively according to 4 subregions, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 4 corresponding respectively mould values of described 4 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 4 mould values ₁..., Mode4 ₁

The concrete computing method of the description vector factor of described first subregion correspondence are as follows:

If I _Gy＞0

|I _Gy|＞|λ ₂I _Gx|λ ₂∈(0，+∞)， $\mathrm{<figure-callout\; id="1"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}1=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gx＜0， $\mathrm{<figure-callout\; id="2"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}2=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

else $\mathrm{Model}4=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

If I _Gy＜0

|I _Gy|＞|λ ₂I _Gx|λ ₂∈(0，+∞)， $\mathrm{Model}3=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gx＜0， $\mathrm{<figure-callout\; id="2"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}2=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

else $\mathrm{Model}4=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

At described first sampled point, calculate described first sampled point correspondence Then, in described first subregion, choose second sampled point, according to described first sampled point correspondence

Computation process, calculate second sampled point correspondence

And the like, calculate all sampled point correspondences in described first subregion

The sampled point correspondence of the Rule of judgment of described Model1 correspondence will be satisfied

Add up and obtain described Model1 value; The sampled point correspondence of the Rule of judgment of described Model2 correspondence will be satisfied Add up and obtain described Model2 value; The sampled point correspondence of the Rule of judgment of described Model3 correspondence will be satisfied

Add up and obtain described Model3 value; The sampled point correspondence of the Rule of judgment of described Model4 correspondence will be satisfied

Adding up obtains described Model4 value, and the description vector factor of described first subregion correspondence is Mode1 ₁..., Mode4 ₁

Described description vector calculation module 93 comprises, second computing module 932, choose successively in form region second sub regions that is used for around point of interest, choosing and set the individual sampled point of quantity, according to the computation process of the description of described first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode4 ₂

16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 4 * 16=64 ties up ₁..., Mode4 ₁, Mode1 ₂..., Mode4 ₂..., Mode1 ₁₆..., Mode4 ₁₆) ₆₄, described description vector v is the pairing description vector of described first point of interest.

The vectorial factor computing module 92 of described description can comprise: 8 subregion computing modules 923, be used for first subregion at described first sampled point place is divided into 8 subregions, the position coordinates of described first sampled point is that (x, y), the derivative coordinate is (I _Gx, I _Gy).According to the horizontal direction derivative and the vertical direction derivative of each sampled point in first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate.Then, the Rule of judgment corresponding respectively according to 8 subregions, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 8 corresponding respectively mould values of described 8 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 8 mould values ₁..., Mode8 ₁

The concrete computing method of the description vector factor of described first sampled point are as follows:

If I _Gx＞0

I _Gy＞I _Gx， $\mathrm{<figure-callout\; id="1"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}1=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

0＜I _Gy＜I _Gx， $\mathrm{<figure-callout\; id="2"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}2=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

-I _Gx＜I _Gy＜0， $\mathrm{Model}3=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gy＜-I _Gx， $\mathrm{Model}4=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

If I _Gx＜0

I _Gy＞-I _Gx， $\mathrm{<figure-callout\; id="5"\; label="Model"\; filenames="HSA00000311734200022.png"\; state="\{\{state\}\}">Model</figure-callout>}5=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

0＜I _Gy＜-I _Gx， $\mathrm{<figure-callout\; id="6"\; label="Model"\; filenames="HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}6=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gx＜I _Gy＜0， $\mathrm{<figure-callout\; id="7"\; label="Model"\; filenames="HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}7=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

I _Gy＜I _Gx， $\mathrm{<figure-callout\; id="8"\; label="Model"\; filenames="HSA00000311734200022.png,HSA00000311734200041.png"\; state="\{\{state\}\}">Model</figure-callout>}8=\mathrm{\&Sigma;}{\left|\right|\&dtri;I(x,y)\left|\right|}_2*w_{x,y}$

At described first sampled point, calculate described first sampled point correspondence

Then, in described first subregion, choose second sampled point, according to described first sampled point correspondence

Computation process, calculate second sampled point correspondence

And the like, calculate all sampled point correspondences in described first subregion

Then, the sampled point correspondence of the Rule of judgment of described Model1 correspondence will be satisfied

Add up and obtain described Model1 value; The sampled point correspondence of the Rule of judgment of described Model2 correspondence will be satisfied

Add up and obtain described Model2 value; The sampled point correspondence of the Rule of judgment of described Model3 correspondence will be satisfied Add up and obtain described Model3 value; The sampled point correspondence of the Rule of judgment of described Model4 correspondence will be satisfied

Add up and obtain described Model4 value; The sampled point correspondence of the Rule of judgment of described Model5 correspondence will be satisfied

Add up and obtain described Model5 value; The sampled point correspondence of the Rule of judgment of described Model6 correspondence will be satisfied

Add up and obtain described Model6 value; The sampled point correspondence of the Rule of judgment of described Model7 correspondence will be satisfied

Add up and obtain described Model7 value; The sampled point correspondence of the Rule of judgment of described Model8 correspondence will be satisfied

Add up and obtain described Model8 value; The description vector factor of described first subregion correspondence is Mode1 ₁..., Mode8 ₁

Described description vector calculation module 93 comprises, the 3rd computing module 933, choose successively in form region second sub regions that is used for around point of interest, choosing and set the individual sampled point of quantity, according to the computation process of the description of described first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode8 ₂

16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 8 * 16=128 ties up ₁..., Mode8 ₁, Mode1 ₂..., Mode8 ₂..., Mode1 ₁₆..., Mode8 ₁₆) ₁₂₈, described description vector v is the pairing description vector of described first point of interest.

One of ordinary skill in the art will appreciate that all or part of flow process that realizes in the foregoing description method, be to instruct relevant hardware to finish by computer program, described program can be stored in the computer read/write memory medium, this program can comprise the flow process as the embodiment of above-mentioned each side method when carrying out.Wherein, described storage medium can be magnetic disc, CD, read-only storage memory body (Read-Only Memory, ROM) or at random store memory body (Random Access Memory, RAM) etc.

In sum, the embodiment of the invention is divided into by first subregion with the sampled point place and sets a quantity subregion, calculates the description vector factor of each sampled point correspondence.Thereby realize obviously having reduced the calculated amount of description vector of the point of interest of image.

Each point of interest in the computing method of the description vector of the point of interest that the embodiment of the invention proposes need carry out big or small and 1 floating-point multiplication of judgement of 2-3 time, and its calculated amount is starkly lower than 8 * 4 floating-point multiplication of SIFT method of the prior art.

The computing method of the description vector of the point of interest that the embodiment of the invention proposes have reasonable versatility and the property distinguished, and it is limited but the performance of describing vector is had the application scenario of higher expectation to go for computing power.For example, the application scenario of mobile phone photograph retrieval.

The above; only for the preferable embodiment of the present invention, but protection scope of the present invention is not limited thereto, and anyly is familiar with those skilled in the art in the technical scope that the present invention discloses; the variation that can expect easily or replacement all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of claim.

Claims (14)

1. the computing method of the description vector of the point of interest of an image is characterized in that, comprising:

In Gauss's metric space at point of interest place, choose form region, described form region is divided into sets the quantity sub regions, in each subregion, choose and set the individual sampled point of quantity, calculate the position coordinates and the derivative coordinate of each sampled point;

Described each subregion is divided into a setting quantity subregion, according to the position coordinates and the derivative coordinate of a described setting quantity pairing computing method of subregional division methods and described each sampled point, calculate the description vector factor of each subregion correspondence;

The description vector factor of all subregion correspondences is carried out comprehensively obtaining the description vector of described point of interest.

2. method according to claim 1 is characterized in that, describedly chooses form region in Gauss's metric space at point of interest place, described form region is divided into sets the quantity sub regions, comprising:

Import pending original image, carry out convolution algorithm, obtain a plurality of metric space images with a plurality of not homoscedastic Gaussian functions and described original image;

According to described metric space image and original image, employing difference of Gaussian yardstick detection method obtains point of interest and the characteristic of correspondence yardstick on Gauss's metric space;

Choose foursquare form region in Gauss's metric space at described point of interest place, the width of described form region is provided with according to the characteristic dimension of point of interest, and described form region is divided into 16 subregions that size is identical.

3. method according to claim 1 is characterized in that, described choosing in each subregion set the individual sampled point of quantity, calculates the position coordinates and the derivative coordinate of each sampled point, comprising:

Choose in each subregion in the form region at first point of interest place and set the individual sampled point of quantity, calculate position coordinates, horizontal direction derivative and the vertical direction derivative of each sampled point;

Horizontal direction derivative and vertical direction derivative and the principal direction that calculates described first point of interest according to described first point of interest;

Angle according to the principal direction correspondence of described first point of interest, described form region is carried out shift transformation, obtain the form region under the new coordinate system, calculate the position coordinates and the derivative coordinate of described each sampled point in the form region under described new coordinate system.

4. according to claim 1 or 2 or 3 described methods, it is characterized in that, described described each subregion is divided into set a quantity subregion, position coordinates and derivative coordinate according to a described setting quantity pairing computing method of subregional division methods and described each sampled point, calculate the description vector factor of each subregion correspondence, comprising:

According to the horizontal direction derivative and the vertical direction derivative of each sampled point in first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate;

Described first subregion is divided into 6 subregions, Rule of judgment according to each subregion correspondence, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 6 corresponding respectively mould values of described 6 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 6 mould values ₁..., Mode6 ₁

5. according to claim 1 or 2 or 3 described methods, it is characterized in that, described first subregion with described first sampled point place is divided into sets a quantity subregion, position coordinates and derivative coordinate according to a described setting quantity pairing computing method of subregional division methods and described first sampled point, calculate the description vector factor of described first sampled point correspondence, comprising:

According to the horizontal direction derivative and the vertical direction derivative of each sampled point in first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate;

Described first subregion is divided into 4 subregions, Rule of judgment according to each subregion correspondence, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 4 corresponding respectively mould values of described 4 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 4 mould values ₁..., Mode4 ₁

6. according to claim 1 or 2 or 3 described methods, it is characterized in that, described first subregion with described first sampled point place is divided into sets a quantity subregion, position coordinates and derivative coordinate according to a described setting quantity pairing computing method of subregional division methods and described first sampled point, calculate the description vector factor of described first sampled point correspondence, comprising:

According to the horizontal direction derivative and the vertical direction derivative of each sampled point in first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate;

Described first subregion is divided into 8 subregions, Rule of judgment according to each subregion correspondence, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 8 corresponding respectively mould values of described 8 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 8 mould values ₁..., Mode8 ₁

7. method according to claim 4 is characterized in that, the described description vector factor with all subregion correspondences carries out comprehensively obtaining the description vector of described point of interest, comprising:

Choose successively in second sub regions in the form region of around point of interest, choosing and set the individual sampled point of quantity, according to the computation process of the description of described first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode6 ₂

Then, 16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 6 * 16=96 ties up ₁..., Mode6 ₁, Mode1 ₂..., Mode6 ₂..., Mode1 ₁₆..., Mode6 ₁₆) ₉₆, described description vector v is the pairing description vector of described first point of interest.

8. method according to claim 5 is characterized in that, the described description vector factor with all subregion correspondences carries out comprehensively obtaining the description vector of described point of interest, comprising:

Choose successively in second sub regions in the form region of around point of interest, choosing and set the individual sampled point of quantity, according to the computation process of the description of described first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode4 ₂,

Then, 16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 4 * 16=64 ties up ₁..., Mode4 ₁, Mode1 ₂..., Mode4 ₂..., Mode1 ₁₆..., Mode4 ₁₆) ₆₄, described description vector v is the pairing description vector of described first point of interest.

9. method according to claim 6 is characterized in that, the described description vector factor with all subregion correspondences carries out comprehensively obtaining the description vector of described point of interest, comprising:

Choose successively in second sub regions in the form region of around point of interest, choosing and set the individual sampled point of quantity, according to the computation process of the description of described first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode8 ₂

16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 8 * 16=128 ties up ₁..., Mode8 ₁, Mode1 ₂..., Mode8 ₂..., Mode1 ₁₆..., Mode8 ₁₆) ₁₂₈, described description vector v is the pairing description vector of described first point of interest.

10. the calculation element of the description vector of the point of interest of an image is characterized in that, comprising:

The coordinate Calculation module; Be used for choosing form region, described form region be divided into set the quantity sub regions, in each subregion, choose and set the individual sampled point of quantity, calculate the position coordinates and the derivative coordinate of each sampled point at Gauss's metric space at point of interest place;

Vectorial factor computing module is described, be used for described each subregion is divided into a setting quantity subregion, according to the position coordinates and the derivative coordinate of a described setting quantity pairing computing method of subregional division methods and described each sampled point, calculate the description vector factor of each subregion correspondence;

Describe the vector calculation module, be used for the description vector factor of all subregion correspondences is carried out comprehensively obtaining the description vector of described point of interest.

11. the calculation element of the description vector of the point of interest of image according to claim 10 is characterized in that, described coordinate Calculation module specifically comprises:

Divide processing module, be used to import pending original image, carry out convolution algorithm, obtain a plurality of metric space images with a plurality of not homoscedastic Gaussian functions and described original image; According to described metric space image and original image, employing difference of Gaussian yardstick detection method obtains point of interest and the characteristic of correspondence yardstick on Gauss's metric space;

Choose foursquare form region in Gauss's metric space at described point of interest place, the width of described form region is provided with according to the characteristic dimension of point of interest, and described form region is divided into 16 subregions that size is identical;

The computing module, be used in each subregion of the form region at first point of interest place, choosing the sampled point of setting quantity, calculate position coordinates, horizontal direction derivative and the vertical direction derivative of each sampled point, the principal direction that obtains described first point of interest according to the horizontal direction derivative and the vertical direction derivative calculations of described first point of interest;

Angle according to the principal direction correspondence of described first point of interest, described form region is carried out shift transformation, obtain the form region under the new coordinate system, calculate the position coordinates and the derivative coordinate of described each sampled point in the form region under described new coordinate system.

12. the calculation element of the description vector of the point of interest of image according to claim 10 is characterized in that:

The vectorial factor computing module of described description comprises: 6 subregion computing modules, be used for horizontal direction derivative and vertical direction derivative according to each sampled point of first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate;

Described first subregion is divided into 6 subregions, Rule of judgment according to each subregion correspondence, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 6 corresponding respectively mould values of described 6 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 6 mould values ₁..., Mode6 ₁

Described description vector calculation module comprises, first computing module, be used in second sub regions of the form region that point of interest is chosen, choosing successively the individual sampled point of setting quantity on every side, according to the computation process of the description of described first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode6 ₂

16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 6 * 16=96 ties up ₁..., Mode6 ₁, Mode1 ₂..., Mode6 ₂..., Mode1 ₁₆..., Mode6 ₁₆) ₉₆, described description vector v is the pairing description vector of described first point of interest.

13. the calculation element of the description vector of the point of interest of image according to claim 10 is characterized in that:

The vectorial factor computing module of described description comprises: 4 subregion computing modules, be used for horizontal direction derivative and vertical direction derivative according to each sampled point of first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate;

Described first subregion is divided into 4 subregions, Rule of judgment according to each subregion correspondence, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 4 corresponding respectively mould values of described 4 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 4 mould values ₁..., Mode4 ₁

Described description vector calculation module comprises, second computing module, choose successively in form region second sub regions that is used for around point of interest, choosing and set the individual sampled point of quantity, according to the computation process of the description of described first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode4 ₂

16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 4 * 16=64 ties up ₁..., Mode4 ₁, Mode1 ₂..., Mode4 ₂..., Mode1 ₁₆..., Mode4 ₁₆) ₆₄, described description vector v is the pairing description vector of described first point of interest.

14. the calculation element of the description vector of the point of interest of image according to claim 10 is characterized in that:

The vectorial factor computing module of described description comprises: 8 subregion computing modules, be used for horizontal direction derivative and vertical direction derivative according to each sampled point of first subregion at first sampled point place, calculating with the horizontal direction derivative of each sampled point and vertical direction derivative is the norm of the mould value of the gradient that constitutes of coordinate;

Described first subregion is divided into 8 subregions, Rule of judgment according to each subregion correspondence, the described norm that will belong to each sampled point correspondence in each subregion is weighted accumulation, obtain 8 corresponding respectively mould values of described 8 subregions, constitute the description vector factor M ode1 of described first subregion correspondence according to described 8 mould values ₁..., Mode8 ₁

Described description vector calculation module comprises, the 3rd computing module, choose successively in form region second sub regions that is used for around point of interest, choosing and set the individual sampled point of quantity, according to the computation process of the description of described first subregion correspondence vector factor, calculate the description vector factor M ode1 of the second sub regions correspondence ₂..., Mode8 ₂

16 sub regions all calculate finish after, the description of all the subregion correspondences vector factor is carried out comprehensively, obtain description vector v=(Mode1 that a 8 * 16=128 ties up ₁..., Mode8 ₁, Mode1 ₂..., Mode8 ₂..., Mode1 ₁₆..., Mode8 ₁₆) ₁₂₈, described description vector v is the pairing description vector of described first point of interest.

Images