A method for extracting image texture features based on annular local ternary patterns

Abstract

The present invention provides an image texture feature extraction method based on a circumferential local three-value pattern, by extracting circumferential triangular pixels at various positions in a grayscale image and calculating the grayscale difference pairs thereof, using a ternary threshold function and constructing a descending, ascending and steady-state structural pattern to describe the fluctuation trend of the grayscale difference pairs, expanding the obtained three-value representation feature vector CLTP into a binary representation ascending vector RCLTP and a descending vector DCLTP, and finally merging the statistical histograms of the RCLTP descriptor and the DCLTP descriptor into a joint histogram as the texture feature of the image to be detected. The texture feature quantity calculated based on the present invention can more intuitively and effectively characterize the image texture state, and ensure that the texture description capability is more robust and stable.

Description

Technical Field

[0001] The invention belongs to the technical field of image processing, and in particular relates to an image texture feature extraction method based on a circumferential local three-value pattern. Background Art

[0002] Image texture is defined as the variation of pixel intensity and direction in local space. Texture analysis plays an important role in many applications, so establishing an effective texture representation method is a key step in many computer vision tasks, including target recognition, visual navigation, image segmentation, image retrieval, scene understanding, industrial inspection, etc. So far, a variety of texture representation and classification methods have been proposed. They include statistical methods, structural methods, model methods, and transform domain-based methods, but each method has its own characteristics and shortcomings. Among them, as a typical representation of statistical methods: Local Binary Pattern (LBP) is widely used and has a far-reaching impact. Many deformed texture features have been proposed to enhance the discriminability of LBP.

[0003] Most existing methods cannot take into account the subtle relationship between the center and the neighborhood. For example, LBP is based on the binary comparison of the center pixel and the neighborhood pixel, completely ignoring the differences between the neighborhood pixels, while its improved feature BGC is based on the binary pixel comparison of the adjacent elements in the neighborhood of the center pixel, completely ignoring the center pixel with discriminant information. As a result, the texture image information obtained is incomplete, and subsequent research will have the problem of missing image information. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide an image texture feature extraction method based on annular local ternary patterns, which is used for completely extracting image texture features.

[0005] The technical solution adopted by the present invention to solve the above technical problems is: a method for extracting image texture features based on annular local three-value patterns, comprising the following steps:

[0006] S1: grayscale preprocessing is performed on the image to be tested to obtain a grayscale image;

[0007] S2: Scan the entire grayscale image through the circumferential triangular structure, extract the circumferential triangular pixels at each position in the grayscale image and calculate the grayscale difference pairs of the triangular structure;

[0008] S3: Using the ternary threshold function and constructing descending, ascending and steady-state structural patterns to describe the fluctuation trend of the grayscale difference pair, the ternary texture feature vector CLTP is obtained;

[0009] S4: decompose the ternary texture feature vector CLTP into two binary texture feature vectors, namely, the ascending vector RCLTP and the descending vector DCLTP represented by binary values;

[0010] S5: Merge the statistical histograms of the ascending vector RCLTP descriptor and the descending vector DCLTP descriptor into a joint histogram as the texture feature of the image to be tested.

[0011] According to the above scheme, in step S2, the specific steps are:

[0012] Take a 3×3 window of each pixel of the entire grayscale image and set the grayscale value of the central pixel of the grayscale image to be p c , the 3×3 square neighborhood of the center pixel is [p0,p1…p6,p7], and the center pixel p c The gray value of the square neighborhood pixel is p i ; For each path in the 3×3 window in the image, the central pixel and two adjacent elements in its neighborhood form a triangular structure in clockwise order, and the neighboring pixels are traversed clockwise to form a circular triangular structure. Each window has 8 triangular structure paths, and the pixel set in the path is defined as:

[0013]

[0014] The grayscale difference pair of each path is calculated as:

[0015]

[0016] Furthermore, in step S3, the specific steps are:

[0017] The ternary value mode is used to process the grayscale difference of pixels; is a unit step function, and the threshold T controls the conversion width to distinguish the three levels. Then the grayscale difference pair of the pixel point processed by the ternary threshold function is:

[0018]

[0019] The three structural forms of descending, ascending and steady state are constructed to accurately describe the fluctuation trend of grayscale difference pairs:

[0020] If pixel p i satisfy:

[0021]

[0022] or

[0023] or

[0024] Then the triple [p c ,p i ,p i-1 ] is the ascending structure "1";

[0025] If pixel p i satisfy:

[0026]

[0027] or

[0028] or

[0029] Then the triple [p c ,p i ,p i-1 ] is the descending structure “-1”;

[0030] If pixel p i satisfy:

[0031]

[0032] or

[0033] or

[0034] Then the triple [p c ,p i ,p i-1 ] is the steady-state structure “0”;

[0035] The ternary threshold function υ(·) for rising, falling and steady state is defined as follows, where the threshold T is set according to the empirical value and the degree of grayscale identifiability:

[0036]

[0037] When υ(·) = 1, it is an ascending structure; when υ(·) = -1, it is a descending structure; when υ(·) = 0, it is a steady-state structure; the CLTP texture feature of each pixel is composed of ternary elements of 8 paths combined together to form an 8-bit ternary number.

[0038] Furthermore, in step S4, the specific steps are:

[0039] Each ternary mode of CLTP is decomposed into the binary eigenvalues ​​of the rising mode RCLTP and the falling mode DCLTP, and the corresponding ternary threshold function υ(·) is divided into the binary threshold functions rυ(·) and dυ(·):

[0040]

[0041]

[0042] Furthermore, in step S5, the specific steps are:

[0043] The binary threshold functions rυ(·) and dυ(·) are used to convert the binary rising mode RCLTP feature value and the falling mode DCLTP feature value of each pixel into the RCLTP feature descriptor f RCLTP (p c ) and DCLTP feature descriptor f DCLTP (p c ):

[0044]

[0045]

[0046] The histogram of the RCLTP descriptor and the histogram of the DCLTP descriptor are calculated separately, and the two are combined together using the concatenation method to form a joint CLTP descriptor histogram:

[0047] h CLTP =h RCLTP ∪h DCLTP .

[0048] A computer storage medium stores a computer program executable by a computer processor, wherein the computer program executes an image texture feature extraction method based on a circumferential local three-value pattern.

[0049] The beneficial effects of the present invention are:

[0050] 1. The present invention discloses an image texture feature extraction method based on circumferential local ternary pattern, which extracts the circumferential local ternary pattern (CLTP) feature of the image, and is more effective for image texture analysis than the traditional LBP and its improved descriptor; and realizes the function of completely extracting image texture features.

[0051] 2. The present invention compares the ternary pixels composed of the central pixel and its adjacent elements in the square neighborhood, taking into account the central pixel and the neighborhood pixels, thereby improving the robustness, discrimination and applicability of the texture feature extraction operator.

[0052] 3. The present invention also introduces the concepts of descending, ascending and steady-state structural patterns, describes the local fluctuation trend and improves the accuracy of image texture description.

[0053] 4. The present invention expands the ternary CLTP into binary RCLTP and DCLTP, and merges the statistical histogram into a joint histogram, so that the texture feature quantity can more intuitively and effectively characterize the image texture condition, ensuring that the texture description capability is more robust and stable. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flow chart of an embodiment of the present invention.

[0055] Figure 2 It is a layout structure diagram of the CLTP descriptor according to an embodiment of the present invention.

[0056] Figure 3 It is a ternary structure diagram of the corresponding decline, rise and steady state in the 3×3 grayscale sub-image of the embodiment of the present invention.

[0057] Figure 4 It is a diagram of the eigenvalue effects corresponding to the LBP, BGC1, BGC2, BGC3, LTP and CLTP operators in the three sub-images of an embodiment of the present invention.

[0058] Figure 5 It is a diagram of RCLTP and DCLTP effects corresponding to the ternary pattern of the CLTP descriptor in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0060] See also Figure 1 The specific steps of the image texture feature extraction method based on the annular local three-value pattern of the present invention are as follows:

[0061] Step 1: grayscale the image to be detected;

[0062] Step 2: Scan the entire image through the circumferential triangular structure and calculate the grayscale difference pairs of the triangular structure.

[0063] In step 2, let x and y be pixel p c The grayscale image function is f(x, y). The 3×3 window is used to traverse the entire image. The central pixel in the window and the two adjacent elements in its neighborhood form a triangular structure. The neighborhood pixels are traversed clockwise to form a circular triangular structure, as shown in Figure 2 , so each window has a total of 8 triangular structure paths, and the pixel set in the path is defined as: where p c is the center pixel. The pixel combinations corresponding to each path are:

[0064] {[p c ,p7,p6],[p c ,p6,p5],[p c ,p5,p4],[p c ,p4,p3],[p c ,p3,p2],[p c ,p2,p1],[p c ,p1,p0],[p c ,p0,p7]}.

[0065] Then calculate the grayscale difference pair for each path:

[0066]

[0067] So each pixel p c There are 8 sets of grayscale difference pairs:

[0068]

[0069] Step 3: Using the ternary threshold function and constructing descending, ascending and steady-state structural patterns, a ternary texture feature vector CLTP is obtained.

[0070] In step 3, the grayscale difference is divided into three levels, and a ternary threshold function is used to process the values ​​of grayscale difference pairs:

[0071]

[0072] The threshold T controls the transition width and is set to 60 to distinguish three levels.

[0073] At the same time, for the comparison between the three-element pixels, in order to more accurately describe the fluctuation trend of the grayscale difference pair, three structural forms are constructed: falling, rising and steady state. i The following conditions are met:

[0074]

[0075] or

[0076] or

[0077] Then this triple [p c ,p i ,p i-1 ] is the ascending structure "1";

[0078] If the conditions are met:

[0079]

[0080] or

[0081] or

[0082] Then this triple [p c ,p i ,p i-1 ] is the descending structure “-1”;

[0083] If the conditions are met:

[0084]

[0085] or

[0086] or

[0087] Then this triple [p c ,p i ,p i-1 ] is the steady-state structure “0”;

[0088] like Figure 3 As shown, specific values ​​are used to show the three-element structure of rising, falling and steady state in the image. These three structural forms can be defined by the three-element threshold function υ(·) of rising, falling and steady state.

[0089]

[0090] The threshold value T = 60. When υ(·) = 1, it is an ascending structure; when υ(·) = -1, it is a descending structure; when υ(·) = 0, it is a steady-state structure.

[0091] The CLTP texture feature of each pixel is composed of 8 paths of three-valued elements combined to form an 8-bit ternary number. In order to better describe the differences in the describable texture features, Figure 4 The characteristic values ​​of LBP, BGC1, BGC2, BGC3, LTP and CLTP operators are calculated on three different sub-images (a), (b) and (c). The results of LBP, BGC1, BGC2 and BGC3 in images (a) and (b) are the same and their differences cannot be distinguished. Although LTP can distinguish images (a) and (b), the results in images (b) and (c) are the same. However, the CLTP operator of the present invention can distinguish images (a), (b) and (c) well.

[0092] Step 4: Decompose the ternary texture feature vector CLTP into two binary texture feature vectors: the ascending feature vector RCLTP and the descending feature vector DCLTP.

[0093] In step 4, in order to fully display the texture fluctuation and at the same time be able to use a simple binary expression, the present invention decomposes the ternary texture feature vector CLTP into two binary eigenvalues ​​of the rising mode RCLTP and the falling mode DCLTP, and the corresponding ternary threshold function υ(·) can be divided into binary threshold functions rυ(·) and dυ(·):

[0094]

[0095]

[0096] like Figure 5 As shown, for a certain pixel point, CLTP=00(-1)(-1)(-1)(-1)(-1)1, then its RCLTP=00000001, and DCLTP=00111110.

[0097] Step 5: Finally, the statistical histograms of the RCLTP feature vector descriptor and the DCLTP feature vector descriptor are merged into a joint histogram as the texture feature of the image to be detected.

[0098] In step 5, the binary feature values ​​of the rising mode RCLTP and the falling mode DCLTP of each pixel are converted into the RCLTP feature descriptor f RCLTP (p c ) and DCLTP feature descriptor f DCLTP (p c ):

[0099]

[0100]

[0101] Figure 5 In the figure, the binary RCLTP and DCLTP are converted to decimal feature descriptors 1 and 62 respectively.

[0102] Then the histograms of the RCLTP and DCLTP descriptors of the entire image are calculated separately and combined together using the concatenation method to form a joint histogram:

[0103] h CLTP =h RCLTP ∪h DCLTP .

[0104] The above embodiments are only used to illustrate the design ideas and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design ideas disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A method for extracting image texture features based on annular local ternary patterns, characterized in that: The following steps are involved: S1: grayscale preprocessing is performed on the image to be tested to obtain a grayscale image; S2: Scan the entire grayscale image through the circumferential triangular structure, extract the circumferential triangular pixels at each position in the grayscale image and calculate the grayscale difference pairs of the triangular structure; the specific steps are: Take a 3×3 window of each pixel of the entire grayscale image and set the grayscale value of the central pixel of the grayscale image to be p c , the 3×3 square neighborhood of the center pixel is [p0,p1…p6,p7], and the center pixel p c The gray value of the square neighborhood pixel is p i ; For each path in the 3×3 window in the image, the central pixel and two adjacent elements in its neighborhood form a triangular structure in clockwise order, and the neighboring pixels are traversed clockwise to form a circular triangular structure. Each window has 8 triangular structure paths, and the pixel set in the path is defined as: The grayscale difference pair of each path is calculated as: S3: Using the ternary threshold function and constructing descending, ascending and steady-state structural patterns to describe the fluctuation trend of the grayscale difference pair, the ternary texture feature vector CLTP is obtained; The specific steps are: The ternary value mode is used to process the grayscale difference of pixels; is a unit step function, and the threshold T controls the conversion width to distinguish the three levels. Then the grayscale difference pair of the pixel point processed by the ternary threshold function is: The three structural forms of descending, ascending and steady state are constructed to accurately describe the fluctuation trend of grayscale difference pairs: If pixel p i satisfy: or or Then the triple [p c ,p i ,p i-1 ] is the ascending structure "1"; If pixel p i satisfy: or or Then the triple [p c ,p i ,p i-1 ] is the descending structure "-1"; If pixel p i satisfy: or or Then the triple [p c ,p i ,p i-1 ] is the steady-state structure "0"; The ternary threshold function υ(·) for rising, falling and steady state is defined as follows, where the threshold T is set according to the empirical value and the degree of grayscale identifiability: When υ(·)=1, it is an ascending structure; when υ(·)=-1, it is a descending structure; when υ(·)=0, it is a steady-state structure; the CLTP texture feature of each pixel is composed of ternary elements of 8 paths combined to form an 8-bit ternary number; S4: decompose the ternary texture feature vector CLTP into two binary texture feature vectors, namely, the ascending vector RCLTP and the descending vector DCLTP represented by binary values; S5: Merge the statistical histograms of the ascending vector RCLTP descriptor and the descending vector DCLTP descriptor into a joint histogram as the texture feature of the image to be tested.

2. The image texture feature extraction method based on annular local ternary patterns according to claim 1 is characterized in that: In the step S4, the specific steps are: Each ternary mode of CLTP is decomposed into the binary eigenvalues ​​of the rising mode RCLTP and the falling mode DCLTP, and the corresponding ternary threshold function υ(·) is divided into the binary threshold functions rυ(·) and dυ(·):

3. The image texture feature extraction method based on annular local ternary patterns according to claim 2 is characterized in that: In the step S5, the specific steps are: The binary threshold functions rυ(·) and dυ(·) are used to convert the binary rising mode RCLTP feature value and the falling mode DCLTP feature value of each pixel into the RCLTP feature descriptor f RCLTP (p c ) and DCLTP feature descriptor f DCLTP (p c ): The histogram of the RCLTP descriptor and the histogram of the DCLTP descriptor are calculated separately, and the two are combined together using the concatenation method to form a joint CLTP descriptor histogram: h CLTP =h RCLTP ∪h DCLTP 。 4. A computer storage medium, characterized in that: A computer program executable by a computer processor is stored therein, and the computer program executes an image texture feature extraction method based on a circumferential local three-value pattern as described in any one of claims 1 to 3.

Images