KR101351132B1 - Image segmentation apparatus and method based on anisotropic wavelet transform - Google Patents

Abstract

An image segmentation apparatus and a method thereof based on an anisotropic wavelet transform are disclosed. The image segmentation apparatus based on the anisotropic wavelet transform according to the present invention includes: a feature extraction unit which extracts image features by using the anisotropic wavelet transform; an object segmentation unit which segments an object inside an image by using a ridgeline segmentation method about the image features which are extracted by the feature extraction unit; and an over segmentation elimination unit which eliminates an over image segment by using a marker control ridgeline segmentation about the object which is segmented by the object segmentation unit. [Reference numerals] (110) Feature extraction unit; (120) Object segmentation unit; (130) Over segmentation elimination unit

Description

Image Segmentation Apparatus and Method based on Anisotropic Wavelet Transform}

The present invention relates to an image segmentation apparatus and a method thereof, and more particularly, to perform image segmentation efficiently by performing segmentation operation in a digital image in consideration of extraction and classification of image characteristics, An image segmentation apparatus based on anisotropic wavelet transform and a method thereof capable of solving the problem of over segmentation.

Image segmentation is a fundamental step in various areas of image processing, such as pattern recognition, scenes, and image interpretation. Such image segmentation is often used as an initial transformation for general image interpretation and understanding. Image segmentation has many applications in many different fields. Examples of image segmentation applications include remote sensing, medical image interpretation and diagnostics, and computer vision.

Image segmentation involves dividing an image into a series of sets having similar characteristics. Characteristics of such an image are generally expressed in texture, gray level intensity or color information, and in form.

Numerous methods have been proposed for efficient image segmentation. Basically, these methods fall into two categories: supervised and unsupervised.

The supervised method is based primarily on prior knowledge relating to the number of apparent patterns. In a non-repetitive approach, image areas with high similarity are automatically grouped to evaluate the number of classes. One of the best known methods is Otsu's thresholding method. This method iterates through all possible thresholds and calculates values for pixel levels across the foreground or background of each side of the threshold angle. The goal is to find a threshold whose sum of foreground or background is minimal. Cluster organization is based on an algorithm published by Agus et al. This algorithm uses a tight relationship between the image thresholding problem and cluster analysis. The choice of threshold is controlled by the measure of cluster similarity in this algorithm.

Meanwhile, Paul et al proposed a texture gradient based on a watershed transform. In this way the texture gradient of the image is extracted by wavelet transform. The marker location algorithm is then used to distinguish between homogeneous or heterogeneous texture regions. Markers of fractional transformations are used to properly distinguish the identification area. In this method, complex wavelet transform is used. However, a disadvantage of complex wavelet transform is to use 2 ^d . Where d represents the dimension of the signal to be converted.

In particular, the general image segmentation methods have different characteristics according to the characteristics of the input image, and the segmentation method used according to the input image is also different. However, even when the image is divided, the image may be divided into one or excessively divided due to the distribution and overlapping degree of adjacent pixel values. Thus, the efficiency of segmentation of the image is deteriorated.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and performs segmentation of an image efficiently by performing segmentation operation in a digital image in consideration of extraction and classification of image characteristics, and problem of over-division of a conventional fractional transformation. It is an object of the present invention to provide an image segmentation apparatus and method based on anisotropic wavelet transformation, which can solve the problem.

An image segmentation apparatus based on anisotropic wavelet transform according to an embodiment of the present invention, the feature extraction unit for extracting an image characteristic using an anisotropic wavelet transform (anisotropic wavelet transform); An object dividing unit for dividing an object in an image by using a fractional dividing method on an image characteristic extracted by the characteristic extracting unit; And an oversplitting remover for removing the oversized image segmentation using the marker control fractional line segmentation for the object divided by the object divider.

The feature extractor applies a discrete transform in discrete space z ² along a determined direction by sampling the pixels located in the selected direction.

The feature extractor samples the grid in a d x d dimension (where d is an integer) instead of a singular integer matrix, and the grid is expressed as follows.

Here, a ₁ , b ₁ , a ₂ , b ₂ ∈ z ² , and the integer lattice Λ is composed of pixels obtained by a linear combination of two linearly independent integer vectors d ₁ and d ₂ .

The object dividing unit assumes an image gradient size as a topographic surface, and uses the fractional line method to form a catch basin representing a segment based on the pixels flowing in a common minimum. Divide.

The over-split removing unit calculates the positions of all the regional minimums with respect to the object divided by the object dividing unit, and based on the calculated position, the binary image of the electric pixels that display the positions of the regional minimums. May be calculated, and the overlapped image segmentation may be removed based on the calculated binary image.

An image segmentation method based on anisotropic wavelet transform for achieving the above object comprises: extracting an image characteristic using an anisotropic wavelet transform; Dividing the object in the image with respect to the image characteristic extracted by the image characteristic extracting step using a fractional line dividing method; And removing the overlapped image segmentation using the marker control fractional line segmentation for the object segmented by the object segmentation step.

The image characteristic extraction step may apply a discrete transform in discrete space z ² along a determined direction by sampling pixels located in the selected direction.

In the image characteristic extraction step, sampling is performed in a grid shape in a d x d dimension (where d is an integer) instead of a singular integer matrix, and the grid is expressed as follows.

Here, a ₁ , b ₁ , a ₂ , b ₂ ∈ z ² , and the integer lattice Λ is composed of pixels obtained by a linear combination of two linearly independent integer vectors d ₁ and d ₂ .

In the object dividing step, the image slope may be assumed to be a topographic surface, and the object may be divided using a fractional line method of forming a catch distribution representing a segment based on pixels that flow in a common minimum.

The over-image segmentation removing step may include calculating binary positions of all regional minimums with respect to the object divided by the object segmenting step, and based on the calculated position, the binary image of the lightning pixels indicating the positions of the regional minimums. binary image), and based on the calculated binary image, the excess image segmentation may be removed.

According to the present invention, segmentation of an image can be efficiently performed by applying directional wavelet contour and boundary details of the image.

Further, according to the present invention, the problem of over division of the conventional fractional transformation is eliminated by using a marker based on the fractional line application.

In addition, according to the present invention, it is possible to secure a unified object by dividing an object overlapped with a background such as a person and a car on a road in dynamic moving object tracking, and to obtain motion information on a specific object during dynamic moving object tracking. In this case, only a specific object can be obtained separately to obtain only motion information of the corresponding object.

1 is a diagram schematically illustrating a configuration of an image segmentation apparatus based on an anisotropic wavelet transform according to an embodiment of the present invention.
2 is a diagram illustrating an example of multi-resolution analysis of an image.
3 is an example of a direction group, where (a) is [0 °, 90 °], (b) is [0 °, 45 °], (c) is [0 °, -45 °], and (d) is [90 °, 45 °], and (e) show the case of [90 °, -45 °].
4 is a flowchart illustrating an image segmentation method based on anisotropic wavelet transform according to an embodiment of the present invention.
5 is a diagram illustrating an example of image segmentation.
6 is a diagram illustrating another example of image segmentation.
7 is a diagram illustrating an example of color image segmentation.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description, a detailed description of known techniques well known to those skilled in the art may be omitted.

In describing the constituent elements of the present invention, the same reference numerals may be given to constituent elements having the same name, and the same reference numerals may be given thereto even though they are different from each other. However, even in such a case, it does not mean that the corresponding component has different functions according to the embodiment, or does not mean that the different components have the same function. It should be judged based on the description of each component in the example.

In the following description of the embodiments of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear.

In addition, in describing the component of this invention, terms, such as 1st, 2nd, A, B, (a), (b), can be used. These terms are intended to distinguish the constituent elements from other constituent elements, and the terms do not limit the nature, order or order of the constituent elements. When a component is described as being "connected", "coupled", or "connected" to another component, the component may be directly connected or connected to the other component, Quot; may be "connected," "coupled," or "connected. &Quot;

1 is a diagram schematically illustrating a configuration of an image segmentation apparatus based on an anisotropic wavelet transform according to an embodiment of the present invention.

Referring to FIG. 1, an image segmentation apparatus 100 based on anisotropic wavelet transform according to an embodiment of the present invention includes a feature extractor 110, an object divider 120, and an oversplitting remover 130. do.

The feature extractor 110 extracts an image feature using an anisotropic wavelet transform.

Multi-resolution wavelet transforms have achieved excellent quality performance for texture analysis. This technique is a one-dimensional, shifted and extended version of the prototype band pass wavelet function Ψ (t) and a shifted version of the low pass scaling function φ (t). Represented by signal x (t). The wavelet function and the scaling function may be expressed as Equation 1 and Equation 2.

[Equation 1]

&Quot; (2) "

f (x) is a one-dimensional function, φ _{j0 , k} (x) and Ψ _{j, k} (t) represent the scaling and wavelet functions, and these three parameters are functions of the discrete variable x.

&Quot; (3) "

&Quot; (4) "

Here, W _φ (j ₀ , k) and W _Ψ (j, k) represent the scaling and wavelet coefficients obtained after the wavelet transform. Since the wavelet transform method is conceptually simple and the computational complexity is low, the standard two-dimensional wavelet transform has been used successfully in recent years. However, the execution of two-dimensional wavelets is limited by the spatial isotropy of the base function. Efficient directivity cannot be achieved because of the structure along the horizontal and vertical directions only.

In order to overcome this problem, the embodiment of the present invention performs the application of the base function in the direction of anisotropy. With the implementation of this method, contours and edges can be represented efficiently.

Next, the directional wavelet transform will be described.

In order to overcome the problem of isotropic representation of an image, an embodiment of the present invention performs application of a basic function in the direction of anisotropy. In order to apply the discrete transform in discrete space z ² along the determined direction, it is necessary to sample the pixels located in the selected direction. This multi-dimensional sampling operation is defined in a lattice shape as shown in FIG. The lattice in the d dimension is expressed by Equation 5 as dxd rather than a singular integer matrix.

&Quot; (5) "

Here, a ₁ , b ₁ , a ₂ , and b ₂ ∈ z ² . The integer lattice Λ consists of pixels obtained from a linear combination of two linearly independent integer vectors d ₁ and d ₂ . Any integer lattice is a lower lattice of discrete space z ² that can be further divided into det (M _Λ ) excesses of lattice Λ. Each surplus is determined by the shift vector S _k for k = 0,1, ... det (M _Λ ) −1. For example, the direction vectors d ₁ , d ₂ , d ₃ , d _4, and d ₅ are defined as {(1,0), (0,1}, {(1,0), (1,1)}, {( 1,0), (-1,1)}, {(0,1), (1,1)} and {(1,0), (-1, -1)}, the direction group 3 and Equation 6 will be selected.

&Quot; (6) "

At scale S = 2j, the generation matrix M _Λ is selected and wavelet transform is performed along the direction specified by the generation matrix. Here, f ₁ , f ₂ , f ₃ , f _4, and f ₅ are coefficients obtained after wavelet transform in the directions of d ₁ , d ₂ , d ₃ , d _4, and d ₅ .

The object dividing unit 120 divides the object in the image with respect to the image feature extracted by the feature extracting unit 110 by using a fractional dividing method.

The fractional method of division is described as follows.

The features extracted by the feature extractor 110 at an inclination of the image are used to obtain a segmented image. In this step, the fractional transformation assumes the tilt size of the image as the topographic surface. The pixels with the highest gradient magnitude coincide with the watershed lines representing the area boundaries. Water on any pixel surrounded by a common fractional line flows below the common local saturation minimum. Common minimum flowing pixels form a catch basin that represents a segment.

The over dividing remover 130 removes the oversized image segment by using the marker control fractional division for the object divided by the object dividing unit 120.

The marker control fraction line division is explained as follows.

Direct application of fractional transformations can cause over splitting due to noise and other local irregularities in gradients. To overcome this problem, the number of allowed areas is limited by including preprocessing steps designed to bring additional knowledge to the segmentation process. To do this, we introduce the concept of markers.

The marker is connected to the elements belonging to the image. Two types of markers can be used for image segmentation: one internal marker and one external marker. The inner marker is associated with the object of interest and the outer marker is associated with the background.

The procedure for selecting markers can be done in two steps. In the first stage, the preprocessing operation is carried out, and the second stage consists of the definition of a series of criteria. Various methods can be used to calculate the internal markers. Most methods include linear filtering, nonlinear filtering and morphological processing.

First, the image is segmented using fractional transformation on a gradient image. The result of the fractional transformation is overdivided due to the large number of regional minimums. Next, the positions of all the regional minimums are calculated and produce a binary image of the foreground pixels indicating the positions of the regional minimums based on the calculated position. The location of most regional minimums is very shallow and represents the detail of heterogeneous minimums. The series of "spots" are deeper by an arbitrary height threshold than their surroundings.

Next, look for pixels that are certain to belong to external markers or background. The background may search for and display pixels exactly in between the internal markers. This operation is performed by calculating another fractional problem. In particular, the fractional transformation of the distance transformation of the internal marker image is calculated. Then, using the given inner markers and outer markers, the tilted image is corrected by using a process called the minimal imposition technique. This technique modifies the gray scale image so that regional minimums occur only at the indicated position. Other pixel values are “pushed up” as needed to remove all other regional minimums. The tilt image is modified by adding the local minimum position of the inner and outer markers.

4 is a flowchart illustrating an image segmentation method based on anisotropic wavelet transform according to an embodiment of the present invention.

1 to 4, the feature extractor 110 extracts an image feature by using an anisotropic wavelet transform (S110).

The object dividing unit 120 divides the object in the image with respect to the image feature extracted by the feature extracting unit 110 by using a fractional line dividing method (S120).

The over-split removing unit 130 removes the over-split image segmentation using the marker control fractional line segmentation on the object divided by the object dividing unit 120 (S130).

The image characteristic extraction and segmentation technique described above was applied to a hand image of 160x160 pixels, as shown in FIG. Fig. 5A shows the original gray scale image. The inclination of this image is as shown in Fig. 5B, and the divided image after the application of the fractional transformation is as shown in Fig. 5C. Finally, the segmented image after application of marker controlled fractional line segmentation is as shown in FIG.

6 also shows the segmentation of the same hand image using the tilt obtained after the wavelet transform.

An image segmentation method based on the anisotropic wavelet transform according to an embodiment of the present invention, and the difference between the wavelets based on the segmentation method are clearly observed after the image obtained from the fractional transformation, as shown in FIGS. 5C and 6C. do. In particular, FIG. 6D shows that the final image obtained after the segmentation-based marker is more efficient than the image obtained in the wavelet transform. This is due to the fact that the tilted image obtained after the anisotropic wavelet transform shows the contour and the border more efficiently and reliably.

7 is a diagram showing an example of color image segmentation, (a) is an original image, (b) is a gray level image, and (c) is a tilted image after directional wavelet transform. In addition, (d) shows the over-divided image after the fractional line transformation, and (e) shows the divided image after the marker control fractional line transformation.

The present invention is not necessarily limited to these embodiments, as all the constituent elements constituting the embodiment of the present invention are described as being combined or operated in one operation. That is, within the scope of the present invention, all of the components may be selectively coupled to one or more of them. In addition, although all of the components may be implemented as one independent hardware, some or all of the components may be selectively combined to perform a part or all of the functions in one or a plurality of hardware. As shown in FIG. In addition, such a computer program may be stored in a computer-readable medium such as a USB memory, a CD disk, a flash memory, etc., and read and executed by a computer, thereby implementing embodiments of the present invention. As the storage medium of the computer program, a magnetic recording medium, an optical recording medium, a carrier wave medium, or the like may be included.

Furthermore, all terms including technical or scientific terms have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs, unless otherwise defined in the Detailed Description. Terms used generally, such as terms defined in a dictionary, should be interpreted to coincide with the contextual meaning of the related art, and shall not be interpreted in an ideal or excessively formal sense unless explicitly defined in the present invention.

The foregoing description is merely illustrative of the technical idea of the present invention and various changes and modifications may be made by those skilled in the art without departing from the essential characteristics of the present invention. In addition, the embodiments disclosed in the present invention are not intended to limit the scope of the present invention but to limit the scope of the technical idea of the present invention. Accordingly, the scope of protection of the present invention should be construed according to the claims, and all technical ideas within the scope of equivalents should be interpreted as being included in the scope of the present invention.

Claims (10)

A feature extraction unit for extracting an image feature using an anisotropic wavelet transform;
An object dividing unit for dividing an object in an image by using a fractional dividing method on an image characteristic extracted by the characteristic extracting unit; And
An oversplitting remover for removing the oversized image segmentation using marker control fractional line segmentation for the object divided by the object divider.
An image splitter based on wavelet transform of the anisotropy comprising a.
The method of claim 1,
The feature extraction unit,
An image segmentation apparatus based on an anisotropic wavelet transform, characterized by sampling pixels located in a selected direction and applying a discrete transform in discrete space z ² along a determined direction.
3. The method of claim 2,
The feature extraction unit,
An image splitter based on an anisotropic wavelet transform, which performs sampling in the form of a lattice in the dxd dimension (where d is an integer) rather than a singular integer matrix, where the lattice is represented by:

Here, a ₁ , b ₁ , a ₂ , b ₂ ∈ z ² , and the integer lattice Λ is composed of pixels obtained by a linear combination of two linearly independent integer vectors d ₁ and d ₂ .
The method of claim 1,
The object divider,
Assuming that the image slope size is a topographic surface, the object is segmented using a fractional line method that forms a catch basin representing a segment based on a common minimum flowing pixel. An image splitter based on wavelet transform of anisotropy.
The method of claim 1,
The over-division removal unit,
Calculate positions of all the regional minimums with respect to the object divided by the object dividing unit, calculate a binary image of the lightning pixels indicating the positions of the regional minimums based on the calculated position, and calculate An image segmentation apparatus based on anisotropic wavelet transform, wherein the image segmentation is removed based on a binary image.
Extracting an image characteristic using an anisotropic wavelet transform;
Dividing the object in the image with respect to the image characteristic extracted by the image characteristic extracting step using a fractional line dividing method; And
Removing excess image segmentation by using marker-controlled fractional segmentation for the object segmented by the object segmentation;
An image segmentation method based on the anisotropic wavelet transform, comprising a.
The method according to claim 6,
The image characteristic extraction step,
An anisotropic wavelet transform based on anisotropic wavelet transform, characterized by sampling pixels located in a selected direction and applying a discrete transform in discrete space z ² along a determined direction.
8. The method of claim 7,
The image characteristic extraction step,
Sampling in the dxd dimension (where d is an integer) rather than a singular integer matrix, where the lattice is represented by the following equation:

Here, a ₁ , b ₁ , a ₂ , b ₂ ∈ z ² , and the integer lattice Λ is composed of pixels obtained by a linear combination of two linearly independent integer vectors d ₁ and d ₂ .
The method according to claim 6,
The object division step,
An image segmentation method based on anisotropic wavelet transformation, assuming that the image slope size is a topographic surface, and the object is segmented using a fractional line method that forms a catch distribution representing a segment based on pixels flowing in a common minimum.
The method according to claim 6,
The over-image segmentation removing step,
Computing the position of all the regional minimums with respect to the object divided by the object dividing step, calculates a binary image of the electric light pixels indicating the position of the regional minimums based on the calculated position, An image segmentation method based on anisotropic wavelet transform, characterized in that it removes the over-split image segmentation based on a binary image.

Images