Infrared edge detection method for weak moving targets based on frame difference local contrast

Abstract

This invention relates to a method for detecting the edge of weak moving infrared targets based on frame difference local contrast. The method identifies the main false alarm sources—high-brightness cloud edges, blind flash noise, and the inter-frame radiometric characteristics differences of the target. It enhances the target edge through three-frame temporal difference and an improved local contrast calculation method, removing false alarm sources such as flash noise, high-brightness background, and background edges. Then, based on the generated saliency map, threshold segmentation is used to obtain the target edge detection result, and finally, the original image is retrieved to obtain the final accurate target location. This invention addresses the difficulty in simultaneously meeting the requirements of detection rate and false alarm rate for weak infrared targets under complex dynamic backgrounds in practical engineering. Experimental results show that the proposed algorithm has high stability, effectively reducing the number of false alarms and improving target detectability while ensuring detection efficiency.

Description

Technical Field

[0001] This invention relates to the field of infrared small target detection technology, specifically to an infrared weak moving target edge detection method based on frame difference local contrast. Background Technology

[0002] The development of infrared technology has led to its widespread application in many fields such as early warning, surveillance, and terminal guidance. Infrared image target detection is one of the most widely applied and pressing problems to be solved. Due to the limitations of the infrared spectrum, compared to visible light images, infrared images have lower resolution and fewer usable features, making targets easily obscured by background clutter and extremely difficult to detect. Furthermore, infrared imaging systems applied in space are affected by factors such as high-altitude cirrus clouds and platform image shift, resulting in drastic background changes and further complicating target detection. Therefore, addressing the detection of weak infrared targets against dynamic and complex backgrounds has always been a research hotspot and a challenging problem.

[0003] Currently, research on infrared image target detection is mostly divided into two categories: single-frame infrared weak target detection methods and multi-frame infrared weak target detection methods. Due to dynamic changes in the background, field-of-view shift, and the uncertainty of target motion, the performance of multi-frame infrared weak target detection techniques is limited. Therefore, current research mainly focuses on single-frame infrared weak target detection, which can be broadly categorized into: filtering-based detection methods, transform domain-based detection methods, matrix sparse low-rank decomposition-based detection methods, and local contrast-based detection methods. However, single-frame target detection methods require the target to possess local saliency, making it difficult to detect weak targets submerged in complex backgrounds. Therefore, designing a highly accurate, low-false-detection, and low-false-alarm-rate weak target detection method in complex dynamic backgrounds is an urgent problem to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes an infrared weak moving target edge detection method based on frame difference local contrast, which addresses the difficulty in simultaneously meeting the requirements of detection rate and false alarm rate for detecting weak infrared targets under complex dynamic backgrounds in practical engineering. This method effectively reduces the number of false alarms and improves target detectability while ensuring detection efficiency.

[0005] The technical solution adopted by this invention to solve the technical problem is as follows:

[0006] The infrared weak moving target edge detection method based on frame difference local contrast of the present invention includes the following steps:

[0007] Step 1: Enhance the target using the three-frame difference method;

[0008] Step 2: Use an improved local contrast calculation method to suppress background clutter and enhance target edges;

[0009] Step 3: Use the target energy calculated using the center window to enhance the target edge;

[0010] Step 4: Perform threshold segmentation on the second feature map, and use the coarse detection results to return to the first differential feature map for target localization, thereby obtaining accurate detection results.

[0011] Furthermore, step 1 specifically includes:

[0012] Calculate the difference D between the second and third frames in three consecutive images. Img3 And the difference D between the second frame and the first frame Img1 The first differential feature map DI for detection in the second frame image is obtained by reconstructing the feature map. Img By replacing spatial noise with temporal noise, the contrast between the target and the background is enhanced.

[0013] Furthermore, step 2 specifically includes:

[0014] Infrared images consist of background, target, and noise. The background exhibits a flat spatial distribution and slow temporal variation, while noise is isolated both spatially and temporally. An improved local contrast algorithm uses a 2×2 sorting filter to perform full-image sorting filtering on the first difference feature map from step one. The median grayscale value is used as the grayscale estimate for the central target window, constructing an intermediate window to isolate the target from the background. Local contrast calculations are performed using the difference between the edge region of a weak infrared target and its neighboring background region, enhancing the target while suppressing background clutter.

[0015] Furthermore, step 3 specifically includes:

[0016] Due to the characteristics of infrared imaging systems, targets appear as diffuse spots and change rapidly over time. By calculating the difference between the maximum and minimum grayscale values ​​in a 2×2 sorting filter, and utilizing the edge diffusion characteristics of the target, noise and abnormal pixels in both the temporal and spatial domains can be removed, background edges can be suppressed, and target edges can be enhanced.

[0017] Furthermore, step 4 specifically involves:

[0018] To avoid interference from moving background edges such as cirrus clouds after differential mapping on target detection, adaptive thresholding is used to remove unstable response points in the image, resulting in the detection of diffuse target edges. The target edge detection results are then used to recalculate the first differential feature map (DI). Img Target localization is performed to obtain accurate detection results.

[0019] The beneficial effects of this invention are as follows: This invention provides an infrared weak moving target edge detection method based on frame difference local contrast. It utilizes frame difference and local contrast to achieve infrared weak target detection, mainly including: converting spatial noise into temporal noise using a three-frame difference method, removing large areas of bright background, and enhancing the contrast between the target and the background. Through an improved local contrast algorithm, it enhances the target by utilizing the target's diffuse speckle characteristics and local saliency on the first difference feature map, eliminating isolated bright pixels and edge responses, and detecting the target's diffuse edges through threshold segmentation. Finally, it returns the first difference feature map and obtains accurate detection results through connected component and neighborhood maximum search.

[0020] Compared with the prior art, the present invention has the following advantages:

[0021] (1) The detection method of the present invention can detect weak infrared targets in the background where the target does not have local saliency. The target detection rate is high, the false alarm rate is low, and the robustness is better.

[0022] (2) The present invention utilizes the three-frame difference method to effectively suppress the interference of bright background and enhance the target.

[0023] (3) This invention utilizes the characteristics of target diffuse spots to perform improved local contrast calculation, effectively suppressing isolated pixels such as background edges and blind flash pixels, and improving the target detection rate and reducing the false alarm rate, thus realizing the detection of extremely weak targets. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall process of the infrared weak moving target edge detection method based on frame difference local contrast of the present invention;

[0025] Figure 2 This is a schematic diagram of an improved local contrast method using a sliding window.

[0026] Figure 3 a in Figure 3 b, Figure 3 c, Figure 3 d, Figure 3 e-①、 Figure 3 e-②、 Figure 3 Figure e-③ is the processing result of the infrared weak moving target edge detection method based on frame difference local contrast of the present invention. Detailed Implementation

[0027] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings:

[0028] This invention discloses an infrared weak moving target edge detection method based on frame difference and local contrast. This method utilizes frame difference and local contrast to detect weak infrared targets, mainly including: converting spatial noise into temporal noise using a three-frame difference method to remove large areas of bright background and enhance the contrast between the target and the background; enhancing the target by utilizing the target's diffuse speckle characteristics and local saliency through an improved local contrast algorithm, eliminating isolated bright pixels and edge responses, and detecting diffuse edges of the target through threshold segmentation; finally, returning the difference map, and obtaining accurate detection results through connected component and neighborhood maximum search. The flowchart is shown below. Figure 1 As shown. This invention can more accurately detect weak infrared targets, effectively improving the detection rate of weak infrared moving targets against moving backgrounds such as cirrus clouds and reducing the false alarm rate while meeting efficiency requirements.

[0029] The specific process of the infrared weak moving target edge detection method based on frame difference local contrast is as follows:

[0030] Step 1: Input three consecutive frames of images, and calculate the difference image D between the second and third frames. Img3 And the difference map D between the second frame and the first frame Img1 The processing result is as follows Figure 3 a, Figure 3 As shown in b.

[0031] Since the target is dynamic in the field of view, while the background and blind pixels are static or slowly changing, the difference map D is more suitable for both bright and dark targets. Img1 and D Img3 In both cases, the target's location is more prominent than the local background in the second frame. Therefore, the first difference feature map (DI) is obtained by calculating the Hadamard product of the two. Img .

[0032] DI Img =D Img3 ×D Img1

[0033] Furthermore, due to the stability of the target radiation, the first differential characteristic map DI Img The target's positional feature value is non-negative in the second frame, but not in other dynamic background locations. Thresholding segmentation can suppress most of the background and noise; the threshold τ is typically set to 3–5.

[0034]

[0035] Step 2: The infrared image includes background, target, and noise. The background exhibits a flat spatial distribution and slow temporal variation, while the noise is isolated both spatially and temporally. After frame difference enhancement and threshold segmentation, most of the background no longer possesses the diffusion characteristics of point diffusion, and is mostly composed of isolated noise points or continuous background edges. An improved local contrast algorithm is proposed to suppress background edges and isolated noise points.

[0036] The local contrast algorithm can be improved by using a 2×2 sorting filter to calculate the grayscale estimate of the central target window and constructing an intermediate window to isolate the target background. Finally, local contrast calculation is performed using the difference between the edge region of the infrared weak target and its neighboring background region, which enhances the target while suppressing background clutter. The processing result is as follows: Figure 3 As shown in c. The specific steps are as follows:

[0037] Step 2.1: The local contrast algorithm proposes using a sliding window to slide pixel-by-pixel across the image from left to right and from top to bottom. Each improved window is divided into three parts, such as... Figure 2 The central area represents the region where the target edge may appear; the middle layer is the target protection layer; and the outermost layer is the background layer. The background layer is divided into eight parts, and the average pixel value of each of the eight neighborhoods is calculated as the background estimate.

[0038]

[0039] Where, N u This indicates the number of pixels contained in the child window. This represents the valid value of the j-th pixel within the i-th sub-window.

[0040] Step 2.2: When the target edge appears in the central region, due to the characteristics of the infrared imaging system, it appears as a diffuse patch, which can be considered to have at least 3 pixel responses. Isolated noise points and background edges after thresholding have fewer response pixels, generally no more than 2. Therefore, a 2×2 central window is used. To suppress isolated bright pixels and continuous background edges, the pixel grayscale values ​​of the second and third grayscale levels are calculated, along with m. T ,exist Figure 2 The values ​​in the middle are pixels A and D, which are used to estimate the gray level of the target edge, and the gray level of the third pixel is restricted to not be 0.

[0041]

[0042] In the formula, I2 represents the gray value of the second gray level pixel, and I3 represents the gray value of the third gray level pixel.

[0043] Step 2.3: Calculate the difference between the target edge sub-window and the background:

[0044] d(T,Bi ) = m T -m i

[0045] Since the background edges are usually continuous, while the target is usually within a small range (this algorithm is limited to the intermediate layer), the MPCM algorithm can effectively suppress continuous backgrounds and calculate the second feature map C.

[0046] C(x mn y mn )=min(d mn (T,B i )*d mn (T,B i+4 ))

[0047] Among them, (x mn y mn ) represents the coordinates of the top-left pixel position in the target area, d mn This indicates the contrast between the target edge sub-window and the background sub-window i at this coordinate.

[0048] Step 3: Due to the characteristics of infrared imaging systems, the target appears as a diffuse patch and changes rapidly over time. Therefore, the sorted gray values ​​in a 2×2 sorting filter can be used to utilize the target's edge diffusion characteristics to remove noise and abnormal pixels in both the temporal and spatial domains, suppress background edges, and enhance the target.

[0049] Improved each sliding window, such as Figure 2 As shown, the discussion will focus on the central window scenario:

[0050] 1. If the central window falls on the center of the target or within a large background edge, the grayscale difference between the four pixels will be small.

[0051] 2. If the central window falls on an isolated noise point, and the number of noise points is generally 1 or 2, then the window has been suppressed in step two;

[0052] 3. If the central window falls on the edge of the target, then there must be a main pixel with the largest gray value and the outermost diffuse pixel with the smallest gray value that is close to the background gray value.

[0053] Therefore, in order to enhance the true target edge and suppress background and noise, the second feature map C is weighted by utilizing the diffusion effect of the target edge.

[0054] CI(x mn ,y mn )=C(x mn y mn )*τ mn

[0055] In the formula, (xmn y mn ) represents the coordinates of the top-left pixel in the target area, τ mn For pixel (x) mn ,y mn The weighted eigenvalues ​​at () are specifically defined as follows:

[0056]

[0057] In the formula, I max,m,n and I min,m,n These represent the maximum and minimum grayscale values ​​in the 2×2 sliding window at pixel positions m and n, respectively; m T,m,n For step two, obtain the target grayscale estimate at pixel positions m,n; B mean,m,n The mean value of the background sliding window centered at points m and n; σ is a protection parameter, which can be set to 10 here. -5 The processing results show that this method can effectively enhance the target edges, as shown in the following figure. Figure 3 As shown in d.

[0058] Step 4: To avoid interference from moving background edges such as cirrus clouds after differentiation on target detection, adaptive thresholding is used to remove unstable response points in the image. Thresholding is applied to the local contrast measurement results of the fused infrared target image to obtain the target detection result (processing result as shown in the image). Figure 3 As shown in e-①). Specifically: if the measurement result CI(x) mn ,y mn If the value of ) is greater than Th, then the pixel is the target to be detected; otherwise, it is not the target.

[0059] The coarse detection results are used to return to the input third frame difference map D. Img3 A neighborhood search is performed on the quadruplex region. If a pixel in the quadruplex region has a gray value greater than the average value of the target location in the saliency map, it is considered a target, and the search is repeated until no target is found (the processing result is as follows). Figure 3 (as shown in e-②).

[0060] For pixels identified as targets, their centroids are calculated for target localization, yielding accurate detection results for subsequent target tracking, target feature processing, etc. (processing results are shown in the image). Figure 3 As shown in e-③, the red box represents the target centroid position, and the yellow box represents the target pixel.

[0061] The above experiments show that the detection method of the present invention achieves a target detection rate of 0.993 when the target is extremely weak, which is relatively high. Moreover, the false alarm rate is only 0.09, which overcomes the technical problem of existing detection methods that cannot effectively detect weak targets that are not significant in the spatial or temporal domains.

[0062] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for edge detection of weak moving targets in infrared based on frame difference local contrast, characterized in that, The method includes: Step 1: Obtain the target image. In three consecutive frames, use the difference method to enhance the contrast of the target to obtain the first difference feature map. Step 2: Suppress background clutter in the image and enhance target edges using an improved local contrast calculation method; the specific process of Step 2 is as follows: use The sorting filter is used for the first difference feature map in step one. The entire image is sorted and filtered. The median gray value is used as the gray value estimate of the central target window. The median window is constructed to isolate the target background. The difference between the edge region of the infrared weak target and its neighboring background region is used to perform local contrast calculation, which enhances the target while suppressing background clutter. Step 3: Enhance the target edge using the target energy calculated using the center window; the specific process of step 3 is as follows: calculate The difference between the maximum and minimum gray values ​​in the sorting filter can be used to eliminate noise and abnormal pixels in the temporal and spatial domains, suppress background edges, and enhance target edges by utilizing the edge diffusion characteristics of the target. Step 4: Adaptive threshold segmentation is used to remove unstable response points in the image to obtain the detection results of the target's diffuse edges. The target edge detection results are then used to return to the differential feature map for target localization to obtain accurate detection results.

2. The infrared weak moving target edge detection method based on frame difference local contrast according to claim 1, characterized in that, The specific process of step one is as follows: Acquire the target image, and calculate the difference between the second and third frames in three consecutive frames. And the difference between the second frame and the first frame The first differential feature map for detection in the second frame image is obtained by reconstructing the feature map. By replacing spatial noise with temporal noise, the contrast between the target and the background is enhanced.

3. The infrared weak moving target edge detection method based on frame difference local contrast according to claim 2, characterized in that, First differential feature map It is obtained through the following formula: 。 4. The infrared weak moving target edge detection method based on frame difference local contrast according to claim 2, characterized in that, In step one, threshold segmentation is used to suppress background and noise. The threshold segmentation is specifically as follows: ， in, The threshold value is used.

5. The infrared weak moving target edge detection method based on frame difference local contrast according to claim 1, characterized in that, Step two includes the following: Step 2.1: Decompose the image into four pixels, namely the upper left, upper right, lower left, and lower right of the image, and slide a sliding window on the image from left to right and from top to bottom pixel by pixel; Each window is divided into three parts: the central area is the region where the target edge may appear, the middle layer is the target protection layer, and the outermost layer is the background layer. The background layer is further divided into eight parts, and the average pixel value of each of the eight neighborhoods is calculated as the background estimate. ； in, This indicates the number of pixels contained in the child window. This represents the valid value of the j-th pixel within the i-th sub-window; Step 2.2: When the target edge appears at the center, there are at least 3 pixel responses. The center window is used to calculate the gray levels of the second and third gray levels of the pixels. As a grayscale estimate of the target edge, and restricting the grayscale of the third grayscale cell to not be 0: ； In the formula, This represents the grayscale value of the second-highest grayscale pixel. This represents the grayscale value of the third-highest pixel. Step 2.3: Calculate the difference between the target edge sub-window and the background: - ； The MPCM algorithm effectively suppresses continuous backgrounds, and the second feature map C is calculated as follows: ； in, This indicates the coordinates of the top-left pixel in the target area. This represents the contrast between the target edge sub-window and the background sub-window i at coordinates (m, n). This represents the contrast between the target edge sub-window and the background sub-window i+4 at coordinates (m,n).

6. The infrared weak moving target edge detection method based on frame difference local contrast according to claim 2, characterized in that, Step three includes the following: making the following judgments regarding the central window: If the center window falls on the center of the target or on the edge of a large background area, the grayscale difference between the four pixels will be small. If the center window falls on an isolated noise point and the noise quantity is 1 or 2, it means that the window has been suppressed in step two. If the central window falls on the edge of the target, then there must be a main pixel with the largest gray value and the outermost diffuse pixel with the smallest gray value that is close to the background gray value. The second feature map C is weighted using the diffusion effect of the target edge: ； In the formula, This indicates the coordinates of the top-left pixel in the target area. For pixels The weighted eigenvalues ​​at a given point are defined as follows: ； In the formula, and These represent the positions at pixel locations m and n, respectively. Maximum and minimum grayscale values ​​in a sliding window; This is the target grayscale estimate obtained at pixel positions m and n in step two; The mean of the background sliding window centered at points m and n; To protect parameters.

Images