Image feature extraction method based on improved Beamlet transformation and Canny operator

Abstract

The invention discloses an image feature extraction method based on improved Beamlet transformation and a Canny operator, and belongs to the technical field of image processing. Comprising the following steps: S1, carrying out graying processing on an input to-be-processed image; s2, determining the optimal segmentation scale of the image; s3, calculating an energy statistical value of each pixel in the image under different scales; s4, selecting an optimal Beamlet base for describing image features based on the energy statistical value; and S5, combining the selected Beamlet base with a Canny operator, and carrying out deep edge feature extraction on the image. According to the method, accurate edge detection can be effectively realized, and important details of the image are reserved. Compared with a traditional Canny operator, the method shows more excellent performance in the feature extraction process, and provides a more efficient and reliable solution for practical application in the field of image processing.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to an image feature extraction method based on improved Beamlet transform and Canny operator. Background Art

[0002] In the field of modern image processing, feature extraction is a key step in image analysis and understanding. Feature extraction can help identify important structures in images for further analysis and application. In the prior art, edge detection algorithms and Beamlet transforms are commonly used to extract image features, such as: The patent publication number is CN109145910A, the publication date is January 4, 2019, and the patent document named "Image Line Feature Extraction Method for Holographic Reproduction" records a technical solution in which the Canny operator is first used to detect the edge of the binary image, and then the Beamlet line detection algorithm based on a single scale is used to extract local line features. The advantage of this is that by combining the Canny operator with the Beamlet transform, the contour edges of Chinese characters in the image can be detected by the Canny operator, and the small edges with good directionality can be detected by the Beamlet. At the same time, the pseudo edges that may be detected during the Canny operator edge detection are removed, and the contour that is as accurate as possible is finally obtained.

[0003] The patent publication number is CN102663395A, the publication date is September 12, 2012, and the patent name is "Line Detection Method Based on Adaptive Multi-Scale Fast Discrete Beamlet Transform". The patent document records a technical solution. In the technical solution, a line detection method based on adaptive multi-scale fast discrete Beamlet transform is proposed. Since the scale of the initial partition block of the AMFDBT algorithm is adaptively set by the area where the target point exists in the image, the target point will be spread all over the image in the most extreme case, which reduces the amount of calculation to a certain extent. At the same time, the initial Beamlet transform easily causes the straight line to be truncated, and the adaptive change of the partition sub-block makes the final detection result of the straight line more complete and less prone to truncation.

[0004] Traditional edge detection algorithms are widely used in image processing due to their good noise suppression ability and high edge positioning accuracy. However, the Canny operator may face the problem of insufficient edge extraction and loss of details in the case of complex backgrounds and variable image features.

[0005] Although the traditional Beamlet transform can effectively process the edge and line features of the image, it still has the problem of how to choose the best segmentation scale. Improper scale selection will lead to the loss of feature information or overlapping blur, thus affecting the effectiveness of image analysis. Therefore, designing an optimized feature extraction method that can combine the advantages of the Beamlet transform and the Canny operator to improve the accuracy of feature extraction is an urgent problem to be solved in this field. Summary of the invention

[0006] The technical problem to be solved by the present invention is: to overcome the deficiencies of the prior art and provide an image feature extraction method based on an improved Beamlet transform and a Canny operator. By combining the selected Beamlet basis with the Canny operator, the method can effectively achieve accurate edge detection and retain important details of the image. Compared with the traditional Canny operator, it shows more superior performance in the feature extraction process, providing a more efficient and reliable solution for practical applications in the field of image processing.

[0007] The technical solution adopted by the present invention to solve the technical problem is: the image feature extraction method based on improved Beamlet transform and Canny operator is characterized by comprising the following steps: Step S1, grayscale processing is performed on the input image to be processed; Step S2, determining the optimal image segmentation scale; Step S3, calculating the energy statistics of each pixel in the image at different scales; Step S4, selecting the best Beamlet basis for describing image features based on the energy statistics; Step S5, combining the selected Beamlet basis with the Canny operator to perform in-depth edge feature extraction on the image.

[0008] Furthermore, step S2 further includes the following steps: Step S2-1, applying a small-scale edge detection operator to perform preliminary edge detection on the grayscale image to obtain preliminary edge features; Step S2-2, obtaining pixels on the edge as pixels to be detected, determining the energy statistics of each pixel at each scale of the image, taking the scale at which the maximum energy statistics value is located as the optimal segmentation scale of the pixel, and obtaining the optimal segmentation scale of each pixel to be detected; Step S2-3: Using the statistical distribution principle, find the optimal scale distribution range of all pixels to be detected and determine the optimal segmentation scale of the entire image.

[0009] Furthermore, the small-scale edge detection operator in step S2-1 includes Roberts, Prewitt or Sobel operator.

[0010] Furthermore, step S3 further includes the following steps: Step S3-1, after determining the optimal segmentation scale, calculate the transformation coefficient of each Beamlet basis based on the scale and perform energy statistics; Step S3-2, base Discrete Beamlet transform; Step S3-3, respectively counting the energy components of the basis in the horizontal and vertical directions.

[0011] Furthermore, in step S3-1, the Beamlet basis is , its energy use Indicated by, Beamlet basis transformation and basis length are respectively and Indicates that the energy statistics calculation method is: ; In step S3-2, Indicates that the set of all Beamlet bases on the image is recorded as , , It is the base The pixels on Yes The gray value of for: ; In step S3-3, the base length is Said that the base The length components in the horizontal and vertical directions are recorded as and , the base The length of is the sum of the vectors in two directions: ; The energy values ​​thus obtained are used for basis selection.

[0012] Furthermore, step S4 further includes the following steps: Step S4-1, determining the optimal basis selection method model; Step S4-2, determining the adjacent endpoint of the base endpoint; Step S4-3, determining the optimal basis according to the optimal basis selection rule.

[0013] Further, in step S4-1, represents the basis with the largest energy, Represents the optimal basis: ; in, is the number of bases; For the base; For the base energy; ; in, is the number of bases with maximum energy; ; In step S4-2, the adjacent endpoints of the base endpoint refer to the number of endpoints of the base that are adjacent to the optimal base endpoint of the adjacent block of the binary block where the base is located, and are recorded as .

[0014] Furthermore, in step S4-3, the optimal basis selection rule is: Rule 1: The bases to be determined have common points; Only the number of adjacent endpoints of the non-public endpoints of the basis needs to be considered, and whether it is the optimal basis is determined according to the number of adjacent endpoints of the non-public endpoints: if the number of adjacent endpoints of two basis endpoints is equal, then both bases are taken as the optimal basis; if the number of adjacent endpoints of two basis endpoints is not equal, then the basis corresponding to the one with the larger number of adjacent endpoints is taken as the optimal basis; Rule 2: If the basis to be determined has no common points; If the number of adjacent endpoints of one end point of two bases is equal, and the number of adjacent endpoints of one end point is unequal, then compare the numbers of adjacent endpoints of the unequal endpoints and take the base corresponding to the larger one as the optimal base; if the number of adjacent endpoints of the endpoints of two bases is equal, then both bases are taken as the optimal bases; if the number of adjacent endpoints of the endpoints of two bases is unequal, judge the optimal base according to the sum or difference of the number of adjacent endpoints of the two endpoints of the base: if the sum of the number of adjacent endpoints of the two endpoints is unequal, take the base containing the larger one as the optimal base; if the sum of the number of adjacent endpoints of the two endpoints is equal, calculate the difference in the number of adjacent endpoints of the two endpoints and take the base containing the smaller one as the optimal base.

[0015] Compared with the prior art, the present invention has the following beneficial effects: In the image feature extraction method based on improved Beamlet transform and Canny operator of the present application, the improved Beamlet structureless algorithm requires a smaller smoothing parameter value during edge extraction and can achieve a more satisfactory continuous effect.

[0016] In the image feature extraction method based on improved Beamlet transform and Canny operator of the present application, the selection of the optimal basis in the method can better match the intrinsic structure of the image and more accurately extract the target features, and can filter out redundant information irrelevant to the target features to avoid false detection or missed detection.

[0017] In the image feature extraction method based on improved Beamlet transform and Canny operator of the present application, the selection of the optimal basis in the method uses fewer basis functions to represent the main features of the image, thereby reducing the data dimension and facilitating storage, transmission and subsequent processing.

[0018] In the image feature extraction method based on improved Beamlet transform and Canny operator of the present application, the selection of the optimal basis in the present method can reduce the computational complexity of feature extraction, avoid redundant calculations of irrelevant scales and directions, and complete the task faster.

[0019] In the image feature extraction method based on improved Beamlet transform and Canny operator of the present application, the selection of the optimal basis in the method can better adapt to noise interference, extract stable features, and suppress the influence of random noise on feature extraction.

[0020] In the image feature extraction method based on improved Beamlet transform and Canny operator of the present application, the selection of the optimal basis in the method can more accurately and comprehensively represent the target features, avoid extracting irrelevant features, improve accuracy, and avoid missing information.

[0021] In the image feature extraction method based on improved Beamlet transform and Canny operator of the present application, the selection of the optimal basis in the method can adapt to a variety of tasks (such as feature extraction, denoising, classification, compression, etc.), reducing the cost of redesigning the algorithm.

[0022] In the image feature extraction method based on improved Beamlet transform and Canny operator of the present application, the selection of the optimal basis in the present method achieves a good balance between performance, efficiency and robustness, and better solves practical problems.

[0023] Through the image feature extraction method based on the improved Beamlet transform and Canny operator, combined with the selected Beamlet basis and Canny operator, this method can effectively achieve accurate edge detection and retain important details of the image. Compared with the traditional Canny operator, it shows better performance in the feature extraction process and provides a more efficient and reliable solution for practical applications in the field of image processing.

[0024] In the image feature extraction method based on improved Beamlet transform and Canny operator of the present application, the input image to be processed is grayed to reduce the computational complexity and improve the efficiency of subsequent processing.

[0025] In the image feature extraction method based on improved Beamlet transform and Canny operator of the present application, the best Beamlet basis is selected based on the calculated energy statistics, and the Beamlet basis is used to describe the image features, thereby improving the accuracy of the features.

[0026] A small-scale edge detection operator is used to perform preliminary edge detection on the grayscale image to obtain preliminary edge features for subsequent analysis; the energy statistics of each pixel in the image at different scales are calculated to determine the optimal segmentation scale, thereby optimizing the feature extraction effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of the image feature extraction method based on the improved Beamlet transform and Canny operator; Figure 2 It is a schematic diagram of adjacent endpoints of the basis of the image feature extraction method based on the improved Beamlet transform and Canny operator; Figure 3 It is the optimal base diagram when the number of adjacent endpoints of two base endpoints is equal; Figure 4 It is the optimal base diagram when the number of adjacent endpoints of two base endpoints is unequal; Figure 5 The optimal base diagram for two bases with equal numbers of adjacent endpoints of one end and unequal numbers of adjacent endpoints of one end; Figure 6 It is the optimal base diagram when the number of adjacent endpoints of two base endpoints is equal; Figure 7 The optimal base diagram is when the numbers of adjacent endpoints of two base endpoints are not equal and the sum of the numbers of adjacent endpoints of two endpoints is not equal; Figure 8 The optimal base diagram is when the numbers of adjacent endpoints of two base endpoints are not equal and the sum of the numbers of adjacent endpoints of two endpoints is equal; Fig. 9 The original Lena image in Example 1 of the image feature extraction method based on the improved Beamlet transform and Canny operator; Fig.10 For Fig. 9 Schematic diagram after improved Beamlet algorithm processing; Fig.11 For Fig. 9 After the improved Beamlet algorithm is processed, the Canny operator parameters are The processing diagram when taking 2; Fig.12 For Fig. 9 Schematic diagram after traditional Beamlet algorithm processing; Fig.13 For Fig. 9 After the improved Beamlet algorithm is processed, the Canny operator parameters are The processing diagram when taking 2; Fig.14 For Fig. 9 After the improved Beamlet algorithm is processed, the Canny operator parameters are The processing diagram for time 3 is shown; Fig.15 It is the original image in Example 2 of the image feature extraction method based on improved Beamlet transform and Canny operator; Fig.16 For Fig.15 Schematic diagram after improved Beamlet algorithm processing; Fig.17 For Fig.15 After the improved Beamlet algorithm is processed, the Canny operator parameters are The processing diagram for time 3 is shown; Fig.18 is the grayscale image of the image; Fig.19 For Fig.15 After the improved Beamlet algorithm is processed, the Canny operator parameters are The processing diagram for time 3 is shown; Fig. 20 For Fig.15 After the improved Beamlet algorithm is processed, the Canny operator parameters are Processing diagram when taking 4. DETAILED DESCRIPTION

[0028] Figures 1 to 20 The best embodiment of the present invention is shown below in conjunction with the attached Figures 1 to 20 The present invention is further described.

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0030] like Figure 1 As shown, an image feature extraction method based on improved Beamlet transform and Canny operator includes the following steps: Step S1, grayscale processing is performed on the input image to be processed; Step S2, determining the optimal image segmentation scale; Step S2 further comprises the following steps: Step S2-1, applying a small-scale edge detection operator to perform preliminary edge detection on the grayscale image to obtain preliminary edge features; The small-scale edge detection operator can be implemented by various well-known means in the art, such as Roberts, Prewitt or Sobel operators.

[0031] Step S2-2, obtaining pixels on the edge as pixels to be detected, determining the energy statistics of each pixel at each scale of the image, taking the scale at which the maximum energy statistics value is located as the optimal segmentation scale of the pixel, and obtaining the optimal segmentation scale of each pixel to be detected; Step S2-3: Using the statistical distribution principle, find the optimal scale distribution range of all pixels to be detected, so as to finally determine the optimal segmentation scale of the entire image.

[0032] Step S3, calculating the energy statistics of each pixel in the image at different scales; Step S3 further comprises the following steps: Step S3-1, after determining the optimal segmentation scale, calculate the transformation coefficient of each Beamlet basis based on the scale and perform energy statistics.

[0033] Beamlet is based on , its energy use Indicated by, Beamlet basis transformation and basis length are respectively and Indicates that the energy statistics calculation method is: .

[0034] Step S3-2, base Discrete Beamlet Transform: use Indicates that the set of all Beamlet bases on the image is recorded as , , It is the base The pixels on Yes The gray value of for: .

[0035] Step S3-3, respectively, statistical basis in the horizontal and vertical directions of the energy components; Base length: Said that the base The length components in the horizontal and vertical directions are recorded as and , the base The length of is the sum of the vectors in two directions: .

[0036] The energy value thus obtained will be used for subsequent basis selection.

[0037] Step S4, selecting the best Beamlet basis for describing image features based on the energy statistics; Step S4 further comprises the following steps: Step S4-1, determining the optimal basis selection method model; use represents the basis with the largest energy, Represents the optimal basis: ; in, is the number of bases; For the base; For the base energy; ; in, is the number of bases with maximum energy; .

[0038] Step S4-2, determining the adjacent endpoint of the base endpoint; The adjacent endpoints of a base endpoint refer to the number of endpoints that are adjacent to the optimal base endpoint of the adjacent block of the binary block where the base is located, denoted as ; Combination Figure 2 There are four binary blocks in the figure. The current binary block is represented by a solid line, and the basis to be judged is , the other three adjacent blocks are represented by dotted lines, and the optimal bases within them are: , and , for the basis in the current block Endpoint , which is based on the adjacent endpoints of adjacent blocks Endpoint Heki Endpoint .

[0039] Step S4-3, determining the optimal basis according to the optimal basis selection rule; According to the continuity of edge features and the fact that discrete Beamlet bases are constructed by interpolation between any two points on the boundary of a binary block, only the two endpoints of the base are considered. An optimal base selection rule combining directional information is proposed. After calculating the energy statistics of all bases in the binary block, if the maximum energy statistics value corresponds to multiple Beamlet bases, it can be processed according to the following rules. The specific optimal base selection rule is: Rule 1: The bases to be determined have common points; In this case, we only need to consider the number of adjacent endpoints of the non-public endpoints of the basis. Determine whether it is the optimal basis based on the number of adjacent endpoints of the non-public endpoints: if the number of adjacent endpoints of the two basis endpoints is equal, then both bases are taken as the optimal basis; if the number of adjacent endpoints of the two basis endpoints is not equal, then the basis corresponding to the larger number of adjacent endpoints is taken as the optimal basis. Figure 3~4 As shown, the circles represent the common endpoints, where Figure 3 The number of adjacent endpoints of the two base endpoints is equal. , All are optimal bases; Figure 4 The number of adjacent endpoints of two base endpoints is not equal. is the optimal basis.

[0040] Specifically, if the base and There are common endpoints. The optimal basis is determined based on the number of adjacent endpoints of the non-common endpoints. The number of adjacent endpoints of the non-common endpoints is recorded as and : ; Rule 2: If the base to be determined has no common points; In this case, if the number of adjacent endpoints of one of the two bases is equal, and the number of adjacent endpoints of one of the two bases is unequal, then compare the numbers of adjacent endpoints of the unequal endpoints, and take the base corresponding to the larger one as the optimal base; if the number of adjacent endpoints of the two bases is equal, then both bases are taken as the optimal bases; if the number of adjacent endpoints of the two bases is unequal, then the optimal base is determined based on the sum or difference of the number of adjacent endpoints of the two bases. If the sum of the number of adjacent endpoints of the two endpoints is unequal, then take the base containing the larger one as the optimal base; if the sum of the number of adjacent endpoints of the two endpoints is equal, then calculate the difference in the number of adjacent endpoints of the two endpoints, and take the base containing the smaller one as the optimal base. Figures 5 to 8 As shown, Figure 5For two bases, the number of adjacent endpoints of one end is equal, but the number of adjacent endpoints of one end is unequal. is the optimal basis; Figure 6 For the case where the number of adjacent endpoints of the two base endpoints is equal, , All are optimal bases; Figure 7 For the case where the number of adjacent endpoints of the two base endpoints is not equal, the sum of the number of adjacent endpoints of the two endpoints is not equal. is the optimal basis; Figure 8 For the case where the numbers of adjacent endpoints of two base endpoints are not equal, the sum of the numbers of adjacent endpoints of the two endpoints is equal. is the optimal basis.

[0041] Specific: If the base and There is no common endpoint, and the number of adjacent endpoints based on the two endpoints is recorded as: , , and , then: Rule 2-1: The number of adjacent endpoints of two bases is equal, but the number of adjacent endpoints of one endpoint is not equal: Rule 2-2: The number of adjacent endpoints of two base endpoints is equal: Rule 2-3: The number of adjacent endpoints of two base endpoints is not equal: ; ; in, Representation base The sum of the adjacent endpoints of the two endpoints of Representation base The difference between the adjacent endpoints of the two endpoints of Similar.

[0042] The selection of the optimal basis can be further constrained as follows: Where: is any positive integer constant including 0.

[0043] Step S5, combining the selected Beamlet basis with the Canny operator to perform in-depth edge feature extraction on the image.

[0044] The following two examples further illustrate the above-mentioned image feature extraction method based on improved Beamlet transform and Canny operator: Example 1: Example 1 uses a 512×512 pixel “Lena” image to verify the effectiveness of the proposed Beamlet algorithm for face contour extraction based on the improved Beamlet transform and Canny operator image feature extraction method.

[0045] Specifically, the threshold is set to 0.5, the image is first processed using the Beamlet algorithm proposed in the image feature extraction method based on the improved Beamlet transform and the Canny operator, and then the Canny operator is used for further processing.

[0046] Specific, Canny operator standard deviation parameters The value is 2; at the same time, the same image is processed using the traditional Beamlet unstructured algorithm, and then further processed using the Canny operator. The Canny operator parameter The values ​​are 2 and 3 respectively.

[0047] Specific, Fig. 9 This is Lena's original picture. Fig.10 This is the processing result of the Beamlet algorithm proposed in this image feature extraction method based on the improved Beamlet transform and Canny operator. Fig.11 After being processed by the Beamlet algorithm proposed in this image feature extraction method based on the improved Beamlet transform and Canny operator, the Canny operator parameters Take the processing result at 2: Fig.12 This is the result of traditional Beamlet algorithm processing. Fig.13 and Fig.14 After being processed by the literature algorithm, the Canny operator parameters Take the processing results at times 2 and 3 respectively.

[0048] Example 2: A real image of 256×256 pixels in the East Lake area of ​​Wuhan was selected to verify the effectiveness of the improved Beamlet algorithm proposed in this paper for surface contour extraction.

[0049] Specifically, the threshold is set to 0.5, the image is first processed using the Beamlet unstructured algorithm proposed in the image feature extraction method based on the improved Beamlet transform and the Canny operator, and then the Canny operator is used for further processing.

[0050] Specific, Canny operator standard deviation parameters The value is 3; at the same time, the grayscale image is directly processed using the Canny operator and the same image is first processed using the Matlab built-in binarization tool, and then further processed using the Canny operator. The Canny operator parameters The values ​​are 3 and 4 respectively.

[0051] Specific, Fig.15 is the original image, Fig.16 The processing results of the Beamlet algorithm proposed in this image feature extraction method based on the improved Beamlet transform and Canny operator are as follows: Fig.17 After the Beamlet algorithm is processed by the image feature extraction method based on the improved Beamlet transform and Canny operator, the Canny operator parameters Take the processing result at 3 o'clock; Fig.18 is the grayscale image of the image, Fig.19 and Fig. 20 is the Canny operator parameter The grayscale image processing results when 3 and 4 are taken.

[0052] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0053] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An image feature extraction method based on improved Beamlet transform and Canny operator, characterized in that: The steps include: Step S1, grayscale processing is performed on the input image to be processed; Step S2, determining the optimal image segmentation scale; Step S3, calculating the energy statistics of each pixel in the image at different scales; Step S4, selecting the best Beamlet basis for describing image features based on the energy statistics; Step S5, combining the selected Beamlet basis with the Canny operator to perform in-depth edge feature extraction on the image.

2. The image feature extraction method based on improved Beamlet transform and Canny operator according to claim 1, characterized in that: Step S2 further comprises the following steps: Step S2-1, applying a small-scale edge detection operator to perform preliminary edge detection on the grayscale image to obtain preliminary edge features; Step S2-2, obtaining pixels on the edge as pixels to be detected, determining the energy statistics of each pixel at each scale of the image, taking the scale at which the maximum energy statistics value is located as the optimal segmentation scale of the pixel, and obtaining the optimal segmentation scale of each pixel to be detected; Step S2-3: Using the statistical distribution principle, find the optimal scale distribution range of all pixels to be detected and determine the optimal segmentation scale of the entire image.

3. The image feature extraction method based on improved Beamlet transform and Canny operator according to claim 2, characterized in that: In step S2-1, the small-scale edge detection operator includes Roberts, Prewitt or Sobel operator.

4. The image feature extraction method based on improved Beamlet transform and Canny operator according to claim 1, characterized in that: Step S3 further includes the following steps: Step S3-1, after determining the optimal segmentation scale, calculate the transformation coefficient of each Beamlet basis based on the scale and perform energy statistics; Step S3-2, base Discrete Beamlet transform; Step S3-3, respectively counting the energy components of the basis in the horizontal and vertical directions.

5. The image feature extraction method based on improved Beamlet transform and Canny operator according to claim 4, characterized in that: In step S3-1, the beamlet basis is , its energy use Indicated by, Beamlet basis transformation and basis length are respectively and Indicates that the energy statistics calculation method is: ； In step S3-2, Indicates that the set of all Beamlet bases on the image is recorded as , , It is the base The pixels on Yes The gray value of for: ； In step S3-3, the base length is Said that the base The length components in the horizontal and vertical directions are recorded as and , the base The length of is the sum of the vectors in two directions: ； The energy values ​​thus obtained are used for basis selection.

6. The image feature extraction method based on improved Beamlet transform and Canny operator according to claim 1, characterized in that: Step S4 further includes the following steps: Step S4-1, determining the optimal basis selection method model; Step S4-2, determining the adjacent endpoint of the base endpoint; Step S4-3, determining the optimal basis according to the optimal basis selection rule.

7. The image feature extraction method based on improved Beamlet transform and Canny operator according to claim 6, characterized in that: In step S4-1, represents the basis with the largest energy, Represents the optimal basis: ； in, is the number of bases; For the base; For the base energy; ； in, is the number of bases with maximum energy; ; In step S4-2, the adjacent endpoints of the base endpoint refer to the number of endpoints of the base that are adjacent to the optimal base endpoint of the adjacent block of the binary block where the base is located, and are recorded as .

8. The image feature extraction method based on improved Beamlet transform and Canny operator according to claim 6, characterized in that: In step S4-3, the optimal basis selection rule is: Rule 1: The bases to be determined have common points; Only the number of adjacent endpoints of the non-public endpoints of the basis needs to be considered, and whether it is the optimal basis is determined according to the number of adjacent endpoints of the non-public endpoints: if the number of adjacent endpoints of two basis endpoints is equal, then both bases are taken as the optimal basis; if the number of adjacent endpoints of two basis endpoints is not equal, then the basis corresponding to the one with the larger number of adjacent endpoints is taken as the optimal basis; Rule 2: If the basis to be determined has no common points; If the number of adjacent endpoints of one end point of two bases is equal, and the number of adjacent endpoints of one end point is unequal, then compare the numbers of adjacent endpoints of the unequal endpoints and take the base corresponding to the larger one as the optimal base; if the number of adjacent endpoints of the endpoints of two bases is equal, then both bases are taken as the optimal bases; if the number of adjacent endpoints of the endpoints of two bases is unequal, judge the optimal base according to the sum or difference of the number of adjacent endpoints of the two endpoints of the base: if the sum of the number of adjacent endpoints of the two endpoints is unequal, take the base containing the larger one as the optimal base; if the sum of the number of adjacent endpoints of the two endpoints is equal, calculate the difference in the number of adjacent endpoints of the two endpoints and take the base containing the smaller one as the optimal base.

Images