An adaptive infrared small target detection method considering neighborhood anisotropy

Abstract

This invention belongs to the field of image processing and target detection technology, specifically relating to an adaptive infrared small target detection method considering neighborhood anisotropy. This invention provides an adaptive infrared small target detection method considering neighborhood anisotropy. It calculates the gray-level responses in different directions around the central region using a template designed in this invention, and subtracts these responses from the gray-level values ​​of the central region to obtain the target saliency responses in different directions. The minimum value among these subtractions is taken as the saliency response of the center point. To adapt to targets of different sizes, this invention designs two types of templates: 7×7 (first template) and 9×9 (second template), which are applied to the above steps respectively. The larger value of the saliency responses at the two different scales is taken as the final saliency response. This fully utilizes the neighborhood gray-level distribution characteristics of small targets in infrared images, and can effectively characterize the target saliency in various directions.

Description

Technical Field

[0001] This invention belongs to the field of image processing and target detection technology, specifically relating to an adaptive infrared small target detection method that considers neighborhood anisotropy. Background Technology

[0002] Infrared detection technology has been widely used in various military and civilian fields due to its excellent all-weather and anti-interference capabilities. However, due to the long imaging distance, complex imaging background, and noise and clutter interference, targets on the imaging plane of infrared detection systems are usually very "small". Commonly used visible light target detection methods often fail in the field of infrared small targets, and methods developed for infrared small target detection are difficult to balance in terms of time overhead and accuracy. That is, existing high-precision detection methods are theoretically complex, such as tensor low-rank decomposition, and have high computational consumption.

[0003] Therefore, there is an urgent need for an accurate and efficient method for detecting small infrared targets. Summary of the Invention

[0004] To address the issue that existing infrared small target detection methods struggle to balance high accuracy and low latency, this invention provides an infrared small target detection method that considers neighborhood anisotropy, resulting in high efficiency and good detection performance.

[0005] The technical solution adopted in this invention is as follows:

[0006] An adaptive infrared small target detection method considering neighborhood anisotropy includes the following steps:

[0007] A: Take the original infrared image containing small targets, and calculate the values ​​based on the designed first template. The grayscale response map is a neighborhood-sized image in 8 directions.

[0008] B: Take the difference between the original infrared image containing the small target and the grayscale response map of the 8 directions obtained in step A, and take the absolute value to obtain the target saliency response map of the 8 directions in the 7×7 neighborhood.

[0009] C: Find the minimum value among the 8 absolute differences obtained in step B pixel by pixel to obtain the target saliency response map of the 7×7 neighborhood;

[0010] D: Take the original infrared image containing small targets, and calculate the values ​​based on the designed second template. Gray-scale response map of 8 directions within a neighborhood size;

[0011] E: Calculate the grayscale mean map of the original infrared image containing the small target with a neighborhood size of 3×3, and subtract it from the grayscale response maps of the 8 directions obtained in step D, and take the absolute value to obtain the target saliency response map of the 8 directions in a 9×9 neighborhood.

[0012] F: Find the minimum value among the 8 absolute differences obtained in step E pixel by pixel to obtain the target saliency response map of the 9×9 neighborhood;

[0013] G: Take the larger value from the results obtained in steps C and F for each pixel to obtain the final target saliency response map;

[0014] H: Normalize and threshold the final target saliency response map to obtain the target segmentation result map.

[0015] After adopting this technical solution, the grayscale response in different directions around the central region is calculated using the template designed in this invention, and the difference between these grayscale responses and the grayscale value of the central region is used to obtain the target saliency response in different directions. The minimum value among these is taken as the saliency response of the center point. It should be noted that this invention is designed to accommodate targets of different sizes. (First template) and (Second Template) The two types of templates are applied to the above steps respectively, and the larger value of the significant responses at the two different scales is taken as the final significant response.

[0016] Furthermore, the method described in step A is... The grayscale response map of the neighborhood size in eight directions can be generated through eight image correlation filtering operations, as follows:

[0017] (1)

[0018] In formula (1): This represents a two-dimensional correlation filtering operation; Represents the raw infrared image containing small targets; Represents the first template One template; Representing the The result of a two-dimensional correlation filtering operation, i.e. Within the neighborhood Gray-scale response map in each direction.

[0019] Further, in step B, the difference between the original infrared image containing the small target and the grayscale response maps in the eight directions obtained in step A is calculated, and the absolute value is taken to obtain the target saliency response map in the eight directions within a 7×7 neighborhood. This is specifically performed through the following operations:

[0020] (2)

[0021] In formula (2): This represents taking the absolute value. Represents the raw infrared image containing small targets. represent Within the neighborhood Gray-scale response map in each direction, represent Within the neighborhood Target saliency response diagram in each direction.

[0022] Furthermore, the minimum value among the eight absolute differences obtained in step B, as described in step C, is used as the target saliency response of the 7×7 neighborhood of the center pixel. This is specifically performed through the following operations:

[0023] (3)

[0024] In formula (3): represent Within the neighborhood Target saliency response map in each direction represent Neighborhood target saliency response map.

[0025] Furthermore, in step D, the original infrared image containing the small target is obtained, and calculations are performed based on the designed second template. The grayscale response map of the neighborhood size in eight directions can be obtained through eight image correlation filtering operations:

[0026] (4)

[0027] In equation (4): This represents a two-dimensional correlation filtering operation. Represents the raw infrared image containing small targets. Represents the second template One template, Representing the The result of a two-dimensional correlation filtering operation, i.e. Within the neighborhood Gray-scale response map in each direction.

[0028] Further, in step E, the calculation of the grayscale mean map of the original infrared image containing the small target with a neighborhood size of 3×3 is performed, and the difference between this and the grayscale response maps of the 8 directions obtained in step D is calculated, and the absolute value is taken to obtain the target saliency response map of the 8 directions within a 9×9 neighborhood. This is specifically performed through the following operations:

[0029] (5) (6)

[0030] In equation (5): This represents a two-dimensional correlation filtering operation. Represents the raw infrared image containing small targets. represent Mean filter template, Represents the grayscale mean;

[0031] In formula (6): This represents taking the absolute value. Represents the grayscale mean image. represent Within the neighborhood Gray-scale response map in each direction, represent Within the neighborhood Target saliency response diagram in each direction.

[0032] Furthermore, the minimum value among the eight absolute differences obtained in step E, as described in step F, is used as the target saliency response of the 9×9 neighborhood of the center pixel. This is specifically performed through the following operations:

[0033] (7)

[0034] In equation (7): represent Within the neighborhood Target saliency response map in each direction represent Neighborhood target saliency response map.

[0035] Furthermore, step G involves taking the larger value from the results obtained in steps C and F pixel by pixel to obtain the final target saliency response map, which is specifically performed through the following operations:

[0036] (8)

[0037] In equation (8): represent Neighborhood target saliency response map represent Neighborhood target saliency response map The response diagram represents the saliency of the final objective.

[0038] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0039] 1. A method for detecting small infrared targets considering neighborhood anisotropy, which calculates the gray-level response in different directions around the central region using a template designed in this invention, and subtracts the gray-level values ​​of the central region to obtain the target saliency response in different directions, taking the minimum value as the saliency response of the center point. It should be noted that this invention is designed to accommodate targets of different sizes. (First template) and (Second Template) The two types of templates are applied to the above steps respectively, and the larger value of the significant responses at the two different scales is taken as the final significant response.

[0040] 2. In this invention, steps A and D calculate the grayscale response in eight directions around the central region based on a specially designed template, making full use of the grayscale distribution characteristics of small targets in infrared images. Combined with steps B and E respectively, the salience of targets in each direction can be well characterized.

[0041] 3. In this invention, steps C and F take the minimum value among the eight absolute differences obtained in steps B and E, respectively. This can retain a large response at real targets that are significant in all directions, while significantly reducing the response at non-targets that are not significant or only significant in a few directions.

[0042] 4. In this invention, step H normalizes and thresholds the target saliency map, which can remove false alarms with low response and retain only the real targets with high response. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is the first template designed in this invention;

[0045] Figure 2 This is the second template designed in this invention;

[0046] Figure 3 This is a flowchart of an infrared small target detection method that considers neighborhood anisotropy;

[0047] Figure 4 This is the original infrared image containing small targets to be detected according to Embodiment 1 of the present invention;

[0048] Figure 5 This is Embodiment 1 of the present invention. Gray-scale response maps in the surrounding eight directions within the neighborhood;

[0049] Figure 6 This is Embodiment 1 of the present invention. Target saliency response map in 8 directions within the neighborhood;

[0050] Figure 7This is Embodiment 1 of the present invention. Neighborhood target saliency response map;

[0051] Figure 8 This is Embodiment 1 of the present invention. Gray-scale response maps in the surrounding eight directions within the neighborhood;

[0052] Figure 9 This is Embodiment 1 of the present invention. Target saliency response map in 8 directions within the neighborhood;

[0053] Figure 10 This is Embodiment 1 of the present invention. Neighborhood target saliency response map;

[0054] Figure 11 This is the final target saliency response diagram of Embodiment 1 of the present invention;

[0055] Figure 12 This is a target segmentation result diagram of Embodiment 1 of the present invention; Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only illustrative of the invention and are not intended to limit the invention; that is, the described embodiments are only some embodiments of the invention, and not all embodiments. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0057] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0058] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0059] An adaptive infrared small target detection method considering neighborhood anisotropy includes the following steps:

[0060] A: Take the original infrared image containing small targets, and calculate the values ​​based on the designed first template. The grayscale response map is a neighborhood-sized image in 8 directions.

[0061] B: Take the difference between the original infrared image containing the small target and the grayscale response map of the 8 directions obtained in step A, and take the absolute value to obtain the target saliency response map of the 8 directions in the 7×7 neighborhood.

[0062] C: Find the minimum value among the 8 absolute differences obtained in step B pixel by pixel to obtain the target saliency response map of the 7×7 neighborhood;

[0063] D: Take the original infrared image containing small targets, and calculate the values ​​based on the designed second template. Gray-scale response map of 8 directions within a neighborhood size;

[0064] E: Calculate the grayscale mean map of the original infrared image containing the small target with a neighborhood size of 3×3, and subtract it from the grayscale response maps of the 8 directions obtained in step D, and take the absolute value to obtain the target saliency response map of the 8 directions in a 9×9 neighborhood.

[0065] F: Find the minimum value among the 8 absolute differences obtained in step E pixel by pixel to obtain the target saliency response map of the 9×9 neighborhood;

[0066] G: Take the larger value from the results obtained in steps C and F for each pixel to obtain the final target saliency response map;

[0067] H: Normalize and threshold the final target saliency response map to obtain the target segmentation result map.

[0068] In this embodiment, the step A described above uses The grayscale response map of the neighborhood size in eight directions can be generated through eight image correlation filtering operations, as follows:

[0069] (1)

[0070] In formula (1): This represents a two-dimensional correlation filtering operation; Represents the raw infrared image containing small targets; Represents the first template One template; Representing the The result of a two-dimensional correlation filtering operation, i.e. Within the neighborhood Gray-scale response map in each direction.

[0071] In this embodiment, step B involves taking the difference between the original infrared image containing the small target and the grayscale response maps in the eight directions obtained in step A, and then taking the absolute value to obtain the target saliency response map in the eight directions within a 7×7 neighborhood. This is specifically performed through the following operations:

[0072] (2)

[0073] In formula (2): This represents taking the absolute value. Represents the raw infrared image containing small targets. represent Within the neighborhood Gray-scale response map in each direction, represent Within the neighborhood Target saliency response diagram in each direction.

[0074] In this embodiment, the minimum value among the eight absolute differences obtained in step B, as described in step C, is used as the target saliency response of the 7×7 neighborhood of the center pixel. This is specifically performed through the following operations:

[0075] (3)

[0076] In formula (3): represent Within the neighborhood Target saliency response map in each direction represent Neighborhood target saliency response map.

[0077] In this embodiment, step D involves taking the original infrared image containing small targets and calculating the value based on the designed second template. The grayscale response map of the neighborhood size in eight directions can be obtained through eight image correlation filtering operations:

[0078] (4)

[0079] In equation (4): This represents a two-dimensional correlation filtering operation. Represents the raw infrared image containing small targets. Represents the second template One template, Representing the The result of a two-dimensional correlation filtering operation, i.e. Within the neighborhood Gray-scale response map in each direction.

[0080] In this embodiment, step E involves calculating the grayscale mean map of the original infrared image containing the small target with a neighborhood size of 3×3, and then subtracting it from the grayscale response maps of the eight directions obtained in step D, taking the absolute value to obtain the target saliency response map of the eight directions within a 9×9 neighborhood. This is specifically performed through the following operations:

[0081] (5) (6)

[0082] In equation (5): This represents a two-dimensional correlation filtering operation. Represents the raw infrared image containing small targets. represent Mean filter template, Represents the grayscale mean;

[0083] In formula (6): This represents taking the absolute value. Represents the grayscale mean image. represent Within the neighborhood Gray-scale response map in each direction, represent Within the neighborhood Target saliency response diagram in each direction.

[0084] In this embodiment, the minimum value among the eight absolute differences obtained in step E, which is calculated pixel by pixel in step F, is used as the target saliency response of the 9×9 neighborhood of the center pixel. This is specifically performed through the following operations:

[0085] (7)

[0086] In equation (7): represent Within the neighborhood Target saliency response map in each direction represent Neighborhood target saliency response map.

[0087] In this embodiment, the step G, which involves taking the larger value from the results obtained in steps C and F pixel by pixel to obtain the final target saliency response map, is specifically performed through the following operations:

[0088] (8)

[0089] In equation (8): represent Neighborhood target saliency response map represent Neighborhood target saliency response map The response diagram represents the saliency of the final objective.

[0090] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0091] Reference Figure 3 The preferred embodiment of the present invention provides an infrared small target detection method considering neighborhood anisotropy, comprising the following steps:

[0092] A. Obtain the original infrared image containing the small target, such as... Figure 4 As shown (small infrared targets are marked with white boxes), the first template designed according to the present invention (see...) Figure 1 ) Calculation The grayscale response maps for the eight directions of the neighborhood size are shown below. Figure 5 As shown;

[0093] (1)

[0094] In formula (1): This represents a two-dimensional correlation filtering operation. Represents the raw infrared image containing small targets. Represents the first in template 1 Templates (i.e., related cores) Representing the The result of a two-dimensional correlation filtering operation (i.e.) Within the neighborhood (Gray response map in each direction).

[0095] B. Take the original infrared image containing the small target and the grayscale response maps of the 8 directions obtained in step A, calculate the difference, and take the absolute value to obtain the target saliency response map of the 8 directions within a 7×7 neighborhood. The result is as follows: Figure 6 As shown;

[0096] (2)

[0097] In formula (2): This represents taking the absolute value. Represents the raw infrared image containing small targets. represent Within the neighborhood Gray-scale response map in each direction, represent Within the neighborhood Target saliency response diagram in each direction.

[0098] C. Calculate the minimum value among the eight absolute differences obtained in step B pixel by pixel to obtain the target saliency response map of the 7×7 neighborhood. The result is as follows: Figure 7 As shown;

[0099] (3)

[0100] In formula (3): represent Within the neighborhood Target saliency response map in each direction represent Neighborhood target saliency response map.

[0101] D. Take the original infrared image containing small targets, and calculate the values ​​based on the designed second template. The grayscale response maps of the neighborhood size in eight directions are shown in the following figures. Figure 8 As shown;

[0102] (4)

[0103] In equation (4): This represents a two-dimensional correlation filtering operation. Represents the raw infrared image containing small targets. Represents the first in template 2 Templates (i.e., related cores) Representing the The result of a two-dimensional correlation filtering operation (i.e.) Within the neighborhood (Gray response map in each direction).

[0104] E. Calculate the grayscale mean map of the original infrared image containing the small target with a neighborhood size of 3×3, and subtract it from the grayscale response maps of the 8 directions obtained in step D, taking the absolute value to obtain the target saliency response map of the 8 directions within a 9×9 neighborhood. The result is as follows. Figure 9 As shown;

[0105] (5) (6)

[0106] In equation (5): This represents a two-dimensional correlation filtering operation. Represents the raw infrared image containing small targets. represent Mean filtering template;

[0107] In formula (6): This represents taking the absolute value. represent Within the neighborhood Gray-scale response map in each direction, represent Within the neighborhood Target saliency response diagram in each direction.

[0108] F. Calculate the minimum value among the eight absolute differences obtained in step E pixel by pixel to obtain the target saliency response map of the 9×9 neighborhood. The result is as follows: Figure 10 As shown;

[0109] (7)

[0110] In equation (7): represent Within the neighborhood Target saliency response map in each direction represent Neighborhood target saliency response map.

[0111] G. Take the larger value from the results obtained in steps C and F for each pixel to obtain the final target saliency response map, as shown below. Figure 11 As shown;

[0112] (8)

[0113] In equation (8): represent Neighborhood target saliency response map represent Neighborhood target saliency response map The response diagram represents the saliency of the final objective.

[0114] H. Normalize and threshold the final target saliency response map to obtain the target segmentation result map, as shown in the figure. Figure 12 As shown.

[0115] The above embodiments demonstrate that by calculating the grayscale response in different directions around the central region of the target area designed by this invention, and subtracting the grayscale value of the central region from the grayscale value, the saliency response of the target in different directions is obtained, and the minimum value is taken as the saliency response of the center point. It should be noted that, in order to adapt to targets of different sizes, this invention designs two types of templates, 7×7 (first template) and 9×9 (second template), which are applied to the above steps respectively, and the larger value of the saliency response at the two different scales is taken as the final saliency response.

[0116] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

Claims

1. An adaptive infrared small target detection method considering neighborhood anisotropy, characterized in that: Includes the following steps: A: Take the original infrared image containing small targets, and calculate the values ​​based on the designed first template. The grayscale response map is a neighborhood-sized image in 8 directions. B: Take the difference between the original infrared image containing the small target and the grayscale response map of the 8 directions obtained in step A, and take the absolute value to obtain the target saliency response map of the 8 directions in the 7×7 neighborhood. C: Find the minimum value among the 8 absolute differences obtained in step B pixel by pixel to obtain the target saliency response map of the 7×7 neighborhood; D: Take the original infrared image containing small targets, and calculate the values ​​based on the designed second template. Gray-scale response map of 8 directions within a neighborhood size; E: Calculate the grayscale mean map of the original infrared image containing the small target with a neighborhood size of 3×3, and subtract it from the grayscale response maps of the 8 directions obtained in step D, and take the absolute value to obtain the target saliency response map of the 8 directions in a 9×9 neighborhood. F: Find the minimum value among the 8 absolute differences obtained in step E pixel by pixel to obtain the target saliency response map of the 9×9 neighborhood; G: Take the larger value from the results obtained in steps C and F for each pixel to obtain the final target saliency response map; H: Normalize and threshold the final target saliency response map to obtain the target segmentation result map; The first template is a 7x7 neighborhood 8-directional template, and the second template is a 9x9 neighborhood 8-directional template. The first and second templates represent different scales of the filter to adapt to the detection of small infrared targets of different scales.

2. The adaptive infrared small target detection method considering neighborhood anisotropy according to claim 1, characterized in that: The method described in step A is The grayscale response map of the neighborhood size in eight directions can be generated through eight image correlation filtering operations, as follows: （1）； In formula (1): This represents a two-dimensional correlation filtering operation; Represents the raw infrared image containing small targets; Represents the first template One template; Representing the The result of a two-dimensional correlation filtering operation, i.e. Within the neighborhood Gray-scale response map in each direction.

3. The adaptive infrared small target detection method considering neighborhood anisotropy according to claim 2, characterized in that: In step B, the difference between the original infrared image containing the small target and the grayscale response maps in the eight directions obtained in step A is calculated, and the absolute value is taken to obtain the target saliency response map in the eight directions within a 7×7 neighborhood. This is specifically performed through the following operations: （2）； In formula (2): This represents taking the absolute value. Represents the raw infrared image containing small targets. represent Within the neighborhood Gray-scale response map in each direction, represent Within the neighborhood Target saliency response diagram in each direction.

4. The adaptive infrared small target detection method considering neighborhood anisotropy according to claim 1, characterized in that: In step C, the minimum value among the eight absolute differences obtained in step B is calculated pixel by pixel and used as the target saliency response of the 7×7 neighborhood of the center pixel. This is specifically performed through the following operations: （3）； In formula (3): represent Within the neighborhood Target saliency response map in each direction represent Neighborhood target saliency response map.

5. The adaptive infrared small target detection method considering neighborhood anisotropy according to claim 1, characterized in that: Step D involves acquiring the original infrared image containing small targets and calculating the values ​​based on the designed second template. The grayscale response map of the neighborhood size in eight directions can be obtained through eight image correlation filtering operations: （4）； In equation (4): This represents a two-dimensional correlation filtering operation. Represents the raw infrared image containing small targets. Represents the second template One template, Representing the The result of a two-dimensional correlation filtering operation, i.e. Within the neighborhood Gray-scale response map in each direction.

6. The adaptive infrared small target detection method considering neighborhood anisotropy according to claim 1, characterized in that: Step E involves calculating the grayscale mean map of the original infrared image containing the small target with a neighborhood size of 3×3, and then subtracting it from the grayscale response maps in the eight directions obtained in step D, taking the absolute value to obtain the target saliency response map in the eight directions within a 9×9 neighborhood. This is specifically performed through the following operations: （5）； （6）； In equation (5): This represents a two-dimensional correlation filtering operation. Represents the raw infrared image containing small targets. represent Mean filter template, Represents the grayscale mean; In formula (6): This represents taking the absolute value. Represents the grayscale mean image. represent Within the neighborhood Gray-scale response map in each direction, represent Within the neighborhood Target saliency response diagram in each direction.

7. The adaptive infrared small target detection method considering neighborhood anisotropy according to claim 1, characterized in that: Step F involves calculating the minimum value among the eight absolute differences obtained in step E, pixel by pixel, to serve as the target saliency response for the 9×9 neighborhood of the center pixel. This is specifically performed through the following steps: （7）； In equation (7): represent Within the neighborhood Target saliency response map in each direction represent Neighborhood target saliency response map.

8. The adaptive infrared small target detection method considering neighborhood anisotropy according to claim 1, characterized in that: In step G, the larger value obtained from the results of steps C and F is taken pixel by pixel to obtain the final target saliency response map. This is specifically performed through the following operations: （8）； In equation (8): represent Neighborhood target saliency response map represent Neighborhood target saliency response map The response diagram represents the saliency of the final objective.

Images