CN107392950A - A kind of across yardstick cost polymerization solid matching method based on weak skin texture detection - Google Patents

Abstract

The invention relates to a stereo matching method based on cross-scale cost aggregation based on weak texture detection, which belongs to the field of computer vision, and particularly relates to a stereo matching method for weak texture images, comprising the following steps: input two color images, and the two color images are respectively For the left image and the right image, use the gradient information of the left image to perform weak texture detection and segmentation on the picture; calculate the matching cost according to the color information and gradient information of the left image and the right image; use the above weak texture detection and segmentation results as a benchmark , perform intra-scale and cross-scale cost aggregation based on Gaussian filtering; use the winner-take-all strategy to calculate the disparity; use left-right consistency detection and a method based on adaptive weights to refine the disparity, and output the disparity image. The present invention achieves the technical purpose of improving the matching accuracy of weak texture areas and obtaining better parallax images under the premise of ensuring the matching accuracy of texture areas.

Description

A Cross-Scale Cost Aggregation Stereo Matching Method Based on Weak Texture Detection

technical field

A cross-scale cost aggregation stereo matching method based on weak texture detection in the present invention belongs to the field of computer vision, and in particular relates to a stereo matching method for weak texture images.

Background technique

Binocular Stereo Vision (Binocular Stereo Vision) is an important form of computer vision. It is based on the principle of parallax and uses imaging equipment to obtain two images of the measured object from different positions, and obtains by calculating the position deviation between corresponding points of the image. A method for 3D geometric information of an object. The quality of 3D information acquisition mainly depends on the accuracy of the disparity map obtained by stereo matching. At present, the problems of stereo matching mainly include external factors such as uneven illumination and overexposure, as well as the characteristics of the picture itself that are difficult for computers to distinguish, such as occlusion, weak texture, and repeated texture. Although a large number of scholars have studied stereo matching for many years, the matching of weak texture areas is still a difficult point in the field of image processing. How to improve the matching accuracy of the weak texture area and obtain a better disparity map under the premise of ensuring the matching accuracy of the texture area is a major problem.

Contents of the invention

The present invention provides a cross-scale cost aggregation stereo matching method based on weak texture detection, which can improve the matching accuracy of weak texture regions and obtain better disparity maps under the premise of ensuring the matching accuracy of texture regions.

The purpose of the present invention is achieved like this:

A cross-scale cost aggregation stereo matching method based on weak texture detection, comprising the following steps:

Step a, input two color images, described two color images are left image and right image respectively, utilize the gradient information of left image to carry out weak texture detection and segmentation to picture;

Step b, calculating the matching cost according to the color information and gradient information of the left image and the right image;

Step c, based on the weak texture detection and segmentation results in step a, perform Gaussian filter-based intra-scale and cross-scale cost aggregation;

Step d, using the winner-take-all strategy to calculate the disparity;

In step e, the parallax is refined by using left-right consistency detection and an adaptive weight-based method, and a parallax image is output.

In the described method of cross-scale cost aggregation stereo matching based on weak texture detection, the weak texture detection and segmentation of the picture in the step a are specifically as follows:

Calculate the gradient value g(x,y) of the pixel point at the left image coordinates (x,y), and compare it with the gradient threshold g _T to determine whether it is a weak texture area. The calculation formula is:

g(x,y)<g _T

In the formula: N(x, y) represents the window centered on the pixel (x, y), M represents the number of pixels in the window, and I(x, y) represents the gray value of the pixel.

In the described method of stereo matching based on cross-scale cost aggregation based on weak texture detection, the calculation of the matching cost in the step b is specifically:

Calculate the matching cost C(p,d) of the stereoscopic color image to the left image I _L and the right image I _R , and its calculation formula is:

C(p,d)=(1-α)·C _AD (p,d)+α·(C _{grad_x} (p,d)+C _{grad_y} (p,d))

In the formula: p is a point in the left image, i=R, G, B represent the three channels of the color image respectively in the formula, T _AD and T _grad represent the truncation threshold of color and gradient respectively; Represents the gradient operator of the picture in the x and y directions respectively; α is the balance factor between the color difference and the gradient difference.

In the described method of cross-scale cost aggregation stereo matching based on weak texture detection, the cost aggregation in step c is specifically:

in, Indicates the matching cost after aggregation, z is the desired optimization target value, W is the Gaussian filter kernel, N is the neighborhood window of pixel p, and q is the neighborhood pixel of p; s∈{0,1,..., S} is the scale parameter, when s=0, C ⁰ represents the original scale matching cost of the image; Represents the aggregation cost of S+1 scales of the image;

In the formula, λ is the regularization factor, use Represents the optimization objective function of formula (11), let have:

Among them, T _high and T _low represent the texture area and weak texture area detected in the previous article respectively; C ₁ and C _1/2 represent the matching cost of the original image scale and one-half scale respectively, and use different sizes of windows for Gaussian After filtering, the final matching cost is obtained after fusion.

In the described method of cross-scale cost aggregation stereo matching based on weak texture detection, the parallax refinement in the step e is specifically:

|D' _L (P)-D' _R (P-D' _R (P))|＜δ

D _LRC (P)=min(D'(PL),D'(PR))

Among them, the left image disparity value D' _L (p) and the right image disparity value D' _R (p-D' _L (p)) of a point p in the disparity map, δ is the threshold of LRC; D'(PL) is The parallax value of the first non-occluded point on the left, D(PR) is the parallax value of the first non-occluded point on the right; WB _pq (I _L ) is a function of the left image, Δc _pq and Δs _pq Respectively, the color difference and spatial Euclidean distance of points p and q in the left image, with are the adjustment parameters of color difference and distance difference respectively; D _w (p) is the filtered image.

Beneficial effect:

The present invention is a cross-scale cost aggregation stereo matching method based on weak texture detection. The present invention proposes a new stereo matching algorithm, which selects an appropriate matching method for whether the image area belongs to weak texture, thereby improving the accuracy of stereo matching. Get better disparity maps.

The stereo matching image pair processed by the algorithm of this embodiment can achieve better results in both the texture area and the weak texture area of the picture, and the error matching rate is reduced (5% lower than that without the weak texture area segmentation algorithm). It shows that the algorithm of this embodiment can improve the matching accuracy of the weak texture area on the premise of ensuring the matching accuracy of the texture area, and obtain a better disparity map.

Description of drawings

Figure 1 is a flowchart of a stereo matching method based on cross-scale cost aggregation based on weak texture detection.

Figure 2 is the Bowling1 disparity map.

Figure 3 is the Lampshade1 disparity map.

Figure 4 is a Monopoly disparity map.

Figure 5 is a Plastic disparity map.

detailed description

The specific embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings.

Specific embodiment one

A cross-scale cost aggregation stereo matching method based on weak texture detection, as shown in Figure 1, includes the following steps:

Step a, input two color images, described two color images are left image and right image respectively, utilize the gradient information of left image to carry out weak texture detection and segmentation to picture;

Step b, calculating the matching cost according to the color information and gradient information of the left image and the right image;

Step c, based on the weak texture detection and segmentation results in step a, perform Gaussian filter-based intra-scale and cross-scale cost aggregation;

Step d, using the winner-take-all strategy to calculate the disparity;

In step e, the parallax is refined by using left-right consistency detection and an adaptive weight-based method, and a parallax image is output.

According to the above steps, select four pictures for comparison, as shown in Figure 2, Figure 3, Figure 4 and Figure 5,

In Figure 2, Figure 2(a) is the left image of the original image of Bowling1; Figure 2(b) is the real disparity map of Bowling1; Figure 2(c) is the weak texture detection result of Bowling1; Figure 2(d) is the final disparity map of Bowling1; Figure 2(e) is the disparity map of Bowling1 without weak texture detection.

In Figure 3, Figure 3(a) is the left image of the original image of Lampshade1; Figure 3(b) is the real disparity map of Lampshade1; Figure 3(c) is the weak texture detection result of Lampshade1; Figure 3(d) is the final disparity map of Lampshade1; Figure 3(e) is the disparity map of Lampshade1 without weak texture detection.

In Figure 4, Figure 4(a) is the left image of the original Monopoly image; Figure 4(b) is the real disparity map of Monopoly; Figure 4(c) is the detection result of Monopoly weak texture; Figure 4(d) is the final disparity map of Monopoly; Figure 4(e) is the disparity map of Monopoly without weak texture detection.

In Figure 5, Figure 5(a) is the left image of the original Plastic image; Figure 5(b) is the real disparity map of Plastic; Figure 5(c) is the weak texture detection result of Plastic; Figure 5(d) is the final disparity map of Plastic; Figure 5(e) is the disparity map of Plastic without weak texture detection.

From the perspective of visual effects, Fig. 2(a) ~ Fig. 2(e), Fig. 3(a) ~ Fig. 3(e), Fig. 4(a) ~ Fig. 4(e), Fig. 5(a) ~ Fig. 5( Disparity maps in e) for subjective evaluation. In (c) of FIG. 2 to FIG. 5 , the black part indicates the detected weak texture area, and the white part indicates the texture area. Comparing the disparity map, it can be seen that in the weak texture area, the result of the disparity map obtained by applying the algorithm of this embodiment is much better than that of the algorithm without weak texture detection.

The method of the present invention is evaluated from the objective evaluation index.

Table 1 shows the error matching rate of the four image pairs with obvious weak texture areas in the middlebury image set by applying two algorithms.

Table 1

It can be seen from Table 1 that in the test results of stereo matching image pairs processed by the two algorithms, the error matching rate of the image processed by the algorithm of this embodiment is reduced by 5% compared with the algorithm without weak texture detection and segmentation. It shows that the algorithm of this embodiment can improve the matching accuracy of the weak texture area on the premise of ensuring the matching accuracy of the texture area, and obtain a better disparity map.

Specific embodiment two

In the described method of cross-scale cost aggregation stereo matching based on weak texture detection, the weak texture detection and segmentation of the picture in the step a are specifically as follows:

Calculate the gradient value g(x,y) of the pixel point at the left image coordinates (x,y), and compare it with the gradient threshold g _T to determine whether it is a weak texture area. The calculation formula is:

g(x,y)<g _T

In the formula: N(x, y) represents the window centered on the pixel (x, y), M represents the number of pixels in the window, and I(x, y) represents the gray value of the pixel.

In the described method of stereo matching based on cross-scale cost aggregation based on weak texture detection, the calculation of the matching cost in the step b is specifically:

Calculate the matching cost C(p,d) of the stereoscopic color image to the left image I _L and the right image I _R , and its calculation formula is:

C(p,d)=(1-α)·C _AD (p,d)+α·(C _{grad_x} (p,d)+C _{grad_y} (p,d))

In the formula: p is a point in the left image, i=R, G, B represent the three channels of the color image respectively in the formula, T _AD and T _grad represent the truncation threshold of color and gradient respectively; Represents the gradient operator of the picture in the x and y directions respectively; α is the balance factor between the color difference and the gradient difference.

In the described method of cross-scale cost aggregation stereo matching based on weak texture detection, the cost aggregation in step c is specifically:

in, Indicates the matching cost after aggregation, z is the desired optimization target value, W is the Gaussian filter kernel, N is the neighborhood window of pixel p, and q is the neighborhood pixel of p; s∈{0,1,..., S} is the scale parameter, when s=0, C ⁰ represents the original scale matching cost of the image; Represents the aggregation cost of S+1 scales of the image;

In the formula, λ is the regularization factor, use Represents the optimization objective function of formula (11), let have:

Among them, T _high and T _low represent the texture area and weak texture area detected in the previous article respectively; C ₁ and C _1/2 represent the matching cost of the original image scale and one-half scale respectively, and use different sizes of windows for Gaussian After filtering, the final matching cost is obtained after fusion.

In the described method of cross-scale cost aggregation stereo matching based on weak texture detection, the parallax refinement in the step e is specifically:

|D' _L (P)-D' _R (P-D' _R (P))|＜δ

D _LRC (P)=min(D'(PL),D'(PR))

Among them, the left image disparity value D' _L (p) and the right image disparity value D' _R (p-D' _L (p)) of a point p in the disparity map, δ is the threshold of LRC; D'(PL) is The parallax value of the first non-occluded point on the left, D′(PR) is the parallax value of the first non-occluded point on the right; WB _pq (I _L ) is the function of the left image, Δc _pq and Δs _pq are the color difference and spatial Euclidean distance between points p and q in the left image respectively, with are the adjustment parameters of color difference and distance difference respectively; D _w (p) is the filtered image.

Claims (5)

1. A cross-scale cost aggregation stereo matching method based on weak texture detection is characterized by comprising the following steps:

step a, inputting two color images, wherein the two color images are a left image and a right image respectively, and performing weak texture detection and segmentation on the images by using gradient information of the left image;

b, calculating matching cost according to the color information and the gradient information of the left image and the right image;

step c, taking the weak texture detection and segmentation result in the step a as a reference, and carrying out inner scale and cross-scale cost aggregation based on Gaussian filtering;

d, calculating the parallax by adopting a win person total-taking strategy;

and e, adopting a left-right consistency detection method and a self-adaptive weight-based method to refine the parallax and output a parallax image.

2. The method for cross-scale cost aggregation stereo matching based on weak texture detection as claimed in claim 1, wherein the weak texture detection and segmentation of the picture in the step a specifically comprises:

calculating the gradient value g (x, y) of the pixel point at the coordinate (x, y) of the left image and the gradient threshold value g_TComparing and judging whether the texture is a weak texture area, wherein the calculation formula is as follows:

g(x,y)＜g_T

<mrow> <mi>g</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mi>M</mi> </mfrac> <mo>&amp;CenterDot;</mo> <munder> <mo>&amp;Sigma;</mo> <mrow> <mo>(</mo> <mi>u</mi> <mo>,</mo> <mi>v</mi> <mo>)</mo> <mo>&amp;Element;</mo> <mi>N</mi> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> </munder> <mrow> <mo>(</mo> <mo>|</mo> <mi>I</mi> <mo>(</mo> <mrow> <mi>u</mi> <mo>,</mo> <mi>v</mi> </mrow> <mo>)</mo> <mo>-</mo> <mi>I</mi> <mo>(</mo> <mrow> <mi>u</mi> <mo>+</mo> <mn>1</mn> <mo>,</mo> <mi>v</mi> </mrow> <mo>)</mo> <mo>+</mo> <mo>|</mo> <mi>I</mi> <mo>(</mo> <mrow> <mi>u</mi> <mo>,</mo> <mi>v</mi> </mrow> <mo>)</mo> <mo>-</mo> <mi>I</mi> <mo>(</mo> <mrow> <mi>u</mi> <mo>,</mo> <mi>v</mi> <mo>+</mo> <mn>1</mn> </mrow> <mo>)</mo> <mo>|</mo> <mo>)</mo> </mrow> </mrow>

in the formula, N (x, y) represents a window with a pixel (x, y) as the center, M represents the number of pixels in the window, and I (x, y) represents the gray scale value of the pixel.

3. The method for cross-scale cost aggregation stereo matching based on weak texture detection as claimed in claim 1, wherein the calculating the matching cost in the step b specifically comprises:

calculating a left image I of a stereoscopic color image pair_LAnd a right image I_RThe matching cost C (p, d) is calculated by the formula:

C(p,d)＝(1-α)·C_AD(p,d)+α·(C_{grad_x}(p,d)+C_{grad_y}(p,d))

<mrow> <msub> <mi>C</mi> <mrow> <mi>A</mi> <mi>D</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>d</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>m</mi> <mi>i</mi> <mi>n</mi> <mrow> <mo>(</mo> <mfrac> <mn>1</mn> <mn>3</mn> </mfrac> <munder> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mi>R</mi> <mo>,</mo> <mi>G</mi> <mo>,</mo> <mi>B</mi> </mrow> </munder> <mo>|</mo> <msubsup> <mi>I</mi> <mi>L</mi> <mi>i</mi> </msubsup> <mo>(</mo> <mi>p</mi> <mo>)</mo> <mo>-</mo> <msubsup> <mi>I</mi> <mi>R</mi> <mi>i</mi> </msubsup> <mo>(</mo> <mrow> <mi>p</mi> <mo>,</mo> <mi>d</mi> </mrow> <mo>)</mo> <mo>|</mo> <mo>,</mo> <msub> <mi>T</mi> <mrow> <mi>A</mi> <mi>D</mi> </mrow> </msub> <mo>)</mo> </mrow> </mrow>

C_{grad_x}(p,d)＝min(|▽_xI_L(p)-▽_xI_R(p,d)|,T_grad)

C_{grad_y}(p,d)＝min(|▽_yI_L(p)-▽_yI_R(p,d)|,T_grad)

in the formula: p is a point in the left image, where i ═ R, G, B represent the three channels of the color image, respectively, and T_ADAnd T_gradRepresenting the cut-off thresholds for color and gradient, respectively, ▽_x、▽_yRepresenting the gradient operators of the picture in the x and y directions, respectively, α is a balance factor between the color difference and the gradient difference.

4. The method for cross-scale cost aggregation stereo matching based on weak texture detection as claimed in claim 1, wherein the cost aggregation in the step c is specifically:

<mrow> <mover> <mi>C</mi> <mo>~</mo> </mover> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>d</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>argmin</mi> <mi>z</mi> </munder> <munder> <mo>&amp;Sigma;</mo> <mrow> <mi>q</mi> <mo>&amp;Element;</mo> <mi>N</mi> </mrow> </munder> <mi>W</mi> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>q</mi> <mo>)</mo> </mrow> <mo>|</mo> <mo>|</mo> <mi>z</mi> <mo>-</mo> <mi>C</mi> <mrow> <mo>(</mo> <mi>q</mi> <mo>,</mo> <mi>d</mi> <mo>)</mo> </mrow> <mo>|</mo> <msup> <mo>|</mo> <mn>2</mn> </msup> </mrow>

<mrow> <mover> <mi>C</mi> <mo>~</mo> </mover> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>d</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <mrow> <mi>q</mi> <mo>&amp;Element;</mo> <mi>N</mi> </mrow> </munder> <mi>W</mi> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>q</mi> <mo>)</mo> </mrow> <mo>&amp;CenterDot;</mo> <mi>C</mi> <mrow> <mo>(</mo> <mi>q</mi> <mo>,</mo> <mi>d</mi> <mo>)</mo> </mrow> </mrow>

<mrow> <mover> <mi>v</mi> <mo>~</mo> </mover> <mo>=</mo> <munder> <mrow> <mi>a</mi> <mi>r</mi> <mi>g</mi> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> <msubsup> <mrow> <mo>{</mo> <msup> <mi>z</mi> <mi>s</mi> </msup> <mo>}</mo> </mrow> <mrow> <mi>s</mi> <mo>=</mo> <mn>0</mn> </mrow> <mi>S</mi> </msubsup> </munder> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>s</mi> <mo>=</mo> <mn>0</mn> </mrow> <mi>S</mi> </munderover> <munder> <mo>&amp;Sigma;</mo> <mrow> <msup> <mi>q</mi> <mi>s</mi> </msup> <mo>&amp;Element;</mo> <msub> <mi>N</mi> <mi>s</mi> </msub> </mrow> </munder> <mi>W</mi> <mrow> <mo>(</mo> <msup> <mi>p</mi> <mi>s</mi> </msup> <mo>,</mo> <msup> <mi>q</mi> <mi>s</mi> </msup> <mo>)</mo> </mrow> <mo>|</mo> <mo>|</mo> <msup> <mi>z</mi> <mi>s</mi> </msup> <mo>-</mo> <msup> <mi>C</mi> <mi>s</mi> </msup> <mrow> <mo>(</mo> <msup> <mi>q</mi> <mi>s</mi> </msup> <mo>,</mo> <msup> <mi>d</mi> <mi>s</mi> </msup> <mo>)</mo> </mrow> <mo>|</mo> <msup> <mo>|</mo> <mn>2</mn> </msup> </mrow>

wherein,representing the aggregated matching cost, wherein z is an expected optimization target value, W is a Gaussian filter kernel, N is a neighborhood window of a pixel p, q is a neighborhood pixel point of p, S ∈ {0, 1.., S } is a scale parameter, and when S is 0, C is the value⁰Representing the original scale matching cost of the image;an aggregate cost representing S +1 scales of the image;

<mrow> <msup> <mover> <mi>C</mi> <mo>~</mo> </mover> <mi>s</mi> </msup> <mrow> <mo>(</mo> <msup> <mi>p</mi> <mi>s</mi> </msup> <mo>,</mo> <msup> <mi>d</mi> <mi>s</mi> </msup> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <mrow> <msup> <mi>q</mi> <mi>s</mi> </msup> <mo>&amp;Element;</mo> <msub> <mi>N</mi> <mi>s</mi> </msub> </mrow> </munder> <mi>W</mi> <mrow> <mo>(</mo> <msup> <mi>p</mi> <mi>s</mi> </msup> <mo>,</mo> <msup> <mi>q</mi> <mi>s</mi> </msup> <mo>)</mo> </mrow> <mo>&amp;CenterDot;</mo> <msup> <mi>C</mi> <mi>s</mi> </msup> <mrow> <mo>(</mo> <msup> <mi>q</mi> <mi>s</mi> </msup> <mo>,</mo> <msup> <mi>d</mi> <mi>s</mi> </msup> <mo>)</mo> </mrow> </mrow>

<mrow> <mover> <mi>v</mi> <mo>~</mo> </mover> <mo>=</mo> <munder> <mrow> <mi>a</mi> <mi>r</mi> <mi>g</mi> <mi>m</mi> <mi>i</mi> <mi>n</mi> </mrow> <msubsup> <mrow> <mo>{</mo> <msup> <mi>z</mi> <mi>s</mi> </msup> <mo>}</mo> </mrow> <mrow> <mi>s</mi> <mo>=</mo> <mn>0</mn> </mrow> <mi>S</mi> </msubsup> </munder> <mrow> <mo>(</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>s</mi> <mo>=</mo> <mn>0</mn> </mrow> <mi>S</mi> </munderover> <munder> <mo>&amp;Sigma;</mo> <mrow> <msup> <mi>q</mi> <mi>s</mi> </msup> <mo>&amp;Element;</mo> <msub> <mi>N</mi> <mi>s</mi> </msub> </mrow> </munder> <mi>W</mi> <mo>(</mo> <mrow> <msup> <mi>p</mi> <mi>s</mi> </msup> <mo>,</mo> <msup> <mi>q</mi> <mi>s</mi> </msup> </mrow> <mo>)</mo> <mo>|</mo> <mo>|</mo> <msup> <mi>z</mi> <mi>s</mi> </msup> <mo>-</mo> <msup> <mi>C</mi> <mi>s</mi> </msup> <mo>(</mo> <mrow> <msup> <mi>q</mi> <mi>s</mi> </msup> <mo>,</mo> <msup> <mi>d</mi> <mi>s</mi> </msup> </mrow> <mo>)</mo> <mo>|</mo> <msup> <mo>|</mo> <mn>2</mn> </msup> <mo>+</mo> <mi>&amp;lambda;</mi> <mo>&amp;CenterDot;</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>s</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>S</mi> </munderover> <mo>|</mo> <mo>|</mo> <msup> <mi>z</mi> <mi>s</mi> </msup> <mo>-</mo> <msup> <mi>z</mi> <mrow> <mi>s</mi> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mo>|</mo> <msup> <mo>|</mo> <mn>2</mn> </msup> <mo>)</mo> </mrow> </mrow>

in the formula, lambda is a regularization factor,by usingAn optimization objective function representing equation (11), andcomprises the following steps:

<mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mrow> <mo>(</mo> <mn>1</mn> <mo>+</mo> <mi>&amp;lambda;</mi> <mo>)</mo> <msup> <mi>z</mi> <mi>s</mi> </msup> <mo>-</mo> <msup> <mi>&amp;lambda;z</mi> <mrow> <mi>s</mi> <mo>+</mo> <mn>1</mn> </mrow> </msup> <mo>=</mo> <msup> <mover> <mi>C</mi> <mo>~</mo> </mover> <mi>s</mi> </msup> <mo>(</mo> <msup> <mi>p</mi> <mi>s</mi> </msup> <mo>,</mo> <msup> <mi>d</mi> <mi>s</mi> </msup> <mo>)</mo> <mo>,</mo> <mi>s</mi> <mo>=</mo> <mn>0</mn> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mo>-</mo> <msup> <mi>&amp;lambda;z</mi> <mrow> <mi>s</mi> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mo>+</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>+</mo> <mn>2</mn> <mi>&amp;lambda;</mi> <mo>)</mo> </mrow> <msup> <mi>z</mi> <mi>s</mi> </msup> <mo>-</mo> <msup> <mi>&amp;lambda;z</mi> <mrow> <mi>s</mi> <mo>+</mo> <mn>1</mn> </mrow> </msup> <mo>=</mo> <msup> <mover> <mi>C</mi> <mo>~</mo> </mover> <mi>s</mi> </msup> <mrow> <mo>(</mo> <msup> <mi>p</mi> <mi>s</mi> </msup> <mo>,</mo> <msup> <mi>d</mi> <mi>s</mi> </msup> <mo>)</mo> </mrow> <mo>,</mo> <mi>s</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mn>2</mn> <mo>,</mo> <mo>...</mo> <mo>,</mo> <mi>S</mi> <mo>-</mo> <mn>1</mn> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mo>-</mo> <msup> <mi>&amp;lambda;z</mi> <mrow> <mi>s</mi> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mo>+</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>+</mo> <mn>2</mn> <mi>&amp;lambda;</mi> <mo>)</mo> </mrow> <msup> <mi>z</mi> <mi>s</mi> </msup> <mo>=</mo> <msup> <mover> <mi>C</mi> <mo>~</mo> </mover> <mi>s</mi> </msup> <mrow> <mo>(</mo> <msup> <mi>p</mi> <mi>s</mi> </msup> <mo>,</mo> <msup> <mi>d</mi> <mi>s</mi> </msup> <mo>)</mo> </mrow> <mo>,</mo> <mi>s</mi> <mo>=</mo> <mi>S</mi> </mrow> </mtd> </mtr> </mtable> </mfenced>

<mrow> <mi>A</mi> <mover> <mi>v</mi> <mo>^</mo> </mover> <mo>=</mo> <mover> <mi>v</mi> <mo>~</mo> </mover> </mrow>

<mrow> <mi>A</mi> <mo>=</mo> <mfenced open = "[" close = "]"> <mtable> <mtr> <mtd> <mrow> <mn>1</mn> <mo>+</mo> <mi>&amp;lambda;</mi> </mrow> </mtd> <mtd> <mrow> <mo>-</mo> <mi>&amp;lambda;</mi> </mrow> </mtd> <mtd> <mrow></mrow> </mtd> <mtd> <mrow></mrow> </mtd> <mtd> <mrow></mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mo>-</mo> <mi>&amp;lambda;</mi> </mrow> </mtd> <mtd> <mrow> <mn>1</mn> <mo>+</mo> <mn>2</mn> <mi>&amp;lambda;</mi> </mrow> </mtd> <mtd> <mrow> <mo>-</mo> <mi>&amp;lambda;</mi> </mrow> </mtd> <mtd> <mrow></mrow> </mtd> <mtd> <mrow></mrow> </mtd> </mtr> <mtr> <mtd> <mrow></mrow> </mtd> <mtd> <mo>...</mo> </mtd> <mtd> <mo>...</mo> </mtd> <mtd> <mrow></mrow> </mtd> <mtd> <mrow></mrow> </mtd> </mtr> <mtr> <mtd> <mrow></mrow> </mtd> <mtd> <mrow></mrow> </mtd> <mtd> <mo>...</mo> </mtd> <mtd> <mo>...</mo> </mtd> <mtd> <mrow></mrow> </mtd> </mtr> <mtr> <mtd> <mrow></mrow> </mtd> <mtd> <mrow></mrow> </mtd> <mtd> <mrow> <mo>-</mo> <mi>&amp;lambda;</mi> </mrow> </mtd> <mtd> <mrow> <mn>1</mn> <mo>+</mo> <mn>2</mn> <mi>&amp;lambda;</mi> </mrow> </mtd> <mtd> <mrow> <mo>-</mo> <mi>&amp;lambda;</mi> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow></mrow> </mtd> <mtd> <mrow></mrow> </mtd> <mtd> <mrow></mrow> </mtd> <mtd> <mrow> <mo>-</mo> <mi>&amp;lambda;</mi> </mrow> </mtd> <mtd> <mrow> <mn>1</mn> <mo>+</mo> <mi>&amp;lambda;</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> </mrow>

<mrow> <mover> <mi>v</mi> <mo>^</mo> </mover> <mo>=</mo> <msup> <mi>A</mi> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mover> <mi>v</mi> <mo>~</mo> </mover> </mrow>

<mrow> <msub> <mover> <mi>C</mi> <mo>~</mo> </mover> <mrow> <mi>f</mi> <mi>i</mi> <mi>n</mi> <mi>a</mi> <mi>l</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>d</mi> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>T</mi> <mrow> <mi>h</mi> <mi>i</mi> <mi>g</mi> <mi>h</mi> </mrow> </msub> <mo>&amp;CenterDot;</mo> <munder> <mi>&amp;Sigma;</mi> <mrow> <mi>q</mi> <mo>&amp;Element;</mo> <mi>N</mi> </mrow> </munder> <msub> <mi>W</mi> <mn>1</mn> </msub> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>q</mi> <mo>)</mo> </mrow> <msub> <mi>C</mi> <mn>1</mn> </msub> <mrow> <mo>(</mo> <mi>q</mi> <mo>,</mo> <mi>d</mi> <mo>)</mo> </mrow> <mo>+</mo> <msub> <mi>T</mi> <mrow> <mi>l</mi> <mi>o</mi> <mi>w</mi> </mrow> </msub> <mo>&amp;CenterDot;</mo> <munder> <mi>&amp;Sigma;</mi> <mrow> <mi>q</mi> <mo>&amp;Element;</mo> <mi>N</mi> </mrow> </munder> <msub> <mi>W</mi> <mn>2</mn> </msub> <mrow> <mo>(</mo> <mi>p</mi> <mo>,</mo> <mi>q</mi> <mo>)</mo> </mrow> <msub> <mi>C</mi> <mrow> <mn>1</mn> <mo>/</mo> <mn>2</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>q</mi> <mo>,</mo> <mi>d</mi> <mo>)</mo> </mrow> </mrow>

wherein, T_highAnd T_lowRespectively representing the texture region and the weak texture region detected in the text; c₁And C_1/2Respectively representing the matching cost of the original image scale and the half scale, carrying out Gaussian filtering by using windows with different sizes, and obtaining the final matching cost after fusion.

5. The method for cross-scale cost aggregation stereo matching based on weak texture detection as claimed in claim 1, wherein the disparity refinement in step e specifically comprises:

|D'_L(P)-D'_R(P-D'_R(P))|＜

D_LRC(P)＝min(D'(PL),D'(PR))

<mrow> <msub> <mi>D</mi> <mi>w</mi> </msub> <mrow> <mo>(</mo> <mi>p</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mo>&amp;Sigma;</mo> <mi>q</mi> </munder> <msub> <mi>WB</mi> <mrow> <mi>p</mi> <mi>q</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>I</mi> <mi>L</mi> </msub> <mo>)</mo> </mrow> <msub> <mi>D</mi> <mrow> <mi>L</mi> <mi>R</mi> <mi>C</mi> </mrow> </msub> <mrow> <mo>(</mo> <mi>q</mi> <mo>)</mo> </mrow> </mrow>

<mrow> <msub> <mi>WB</mi> <mrow> <mi>p</mi> <mi>q</mi> </mrow> </msub> <mo>=</mo> <mi>exp</mi> <mrow> <mo>(</mo> <mo>-</mo> <mfrac> <mrow> <msub> <mi>&amp;Delta;c</mi> <mrow> <mi>p</mi> <mi>q</mi> </mrow> </msub> </mrow> <msubsup> <mi>&amp;sigma;</mi> <mi>c</mi> <mn>2</mn> </msubsup> </mfrac> <mo>+</mo> <mfrac> <mrow> <msub> <mi>&amp;Delta;s</mi> <mrow> <mi>p</mi> <mi>q</mi> </mrow> </msub> </mrow> <msubsup> <mi>&amp;sigma;</mi> <mi>s</mi> <mn>2</mn> </msubsup> </mfrac> <mo>)</mo> </mrow> </mrow>

wherein, the left chart parallax value D 'of one point p in the parallax chart'_L(p) and Right View disparity value D'_R(p-D'_L(p)), a threshold for LRC; d '(PL) is the disparity value for the first non-occluded point on the left, and D' (PR) is the disparity value for the first non-occluded point on the right; WB (wideband weight division multiple Access)_pq(I_L) As a function of the left image, Δ c_pqAnd Δ s_pqRespectively the color difference and the spatial euclidean distance of points p and q in the left image,andadjustment parameters of color difference and distance difference respectively; d_w(p) a filtered image.