Four-layer window-based local contrast infrared small target detection method

Abstract

The present application relates to a method for detecting small infrared targets based on local contrast of four-layer window. The purpose of the present application is to obtain a saliency map by calculating local contrast of original infrared image, and to remove noise points from the saliency map to realize infrared small target detection. Fig. 1 in the description is a flow chart for realizing the present application.

Description

Technical fields:

[0001] This invention belongs to the field of image processing. Specifically, it achieves infrared small target detection by calculating the local contrast of an image to obtain a saliency map and then performing noise reduction processing on the saliency map. Background technology:

[0002] Infrared small target detection algorithms based on single-frame images mainly include filtering-based methods, sparse low-rank matrix reconstruction-based methods, and visual saliency-based methods. When sufficient prior information is available, filtering-based methods exhibit good detection performance; however, obtaining prior information is extremely difficult in practical applications, thus limiting the applicability of these algorithms. Sparse reconstruction-based detection algorithms detect small targets mathematically, but they involve significant computational costs and generally have low detection rates in complex environments. Visual saliency-based detection algorithms not only enhance the target's appearance but also effectively suppress background noise, particularly cloud edge regions. Furthermore, these algorithms can be parallelized, making them valuable for practical applications.

[0003] Analysis of small targets and background in infrared images reveals that, while small targets may not achieve maximum global grayscale in complex scenes, they can still achieve local saliency. The Local Contrast Metric (LCM) algorithm based on HVS, proposed by Chen et al., is a classic detection method employing a contrast mechanism. It enhances the local contrast of the target by utilizing local extrema and means. However, when facing cluttered backgrounds, the target contrast may be lower than the contrast of isolated noise, resulting in an excessively high false alarm rate. Infrared small target detection is of significant value to my country's security, making research on infrared small target detection algorithms crucial.

[0004] The grayscale features of small targets can represent the characteristics of the grayscale pixel distribution within the target. Interference, background, and high-brightness noise differ from the grayscale features of small targets. This invention analyzes the background, interference, and small target features to calculate a local saliency map of the image. Without fusing other information, the analysis results are used to denoise the saliency map, thereby achieving the detection of infrared small targets.

[0005] This invention is a method for detecting small targets by analyzing the grayscale features of small targets in the original infrared image and saliency map, based on the differences in features between interference, background, high-brightness noise and small targets. Summary of the Invention:

[0006] The purpose of this invention is to calculate the saliency map by using a four-layer sliding window to perform local contrast calculation on the original infrared image, then use median filtering and mean filtering to denoise the saliency map, and finally use adaptive thresholding to locate small targets, thereby achieving the purpose of infrared small target detection.

[0007] The technical solution adopted in this invention is:

[0008] Step 1: Read the original infrared image I0. The size of the image is M×N, where M is the width of image I0 in pixels and N is the height of image I0 in pixels.

[0009] Step 2: Design a sliding window to scan the image I0. The sliding window consists of 49 local blocks arranged in 7×7. Each local block consists of 3×3 pixels. The local blocks in the sliding window are numbered as shown in matrix C in equation (1). The block with the number T is called the T region, the block with the number G is called the G region, the block with the number H is called the H region, and the block with the number B is called the B region.

[0010]

[0011] Step 3: Calculate the average value of the three largest pixel grayscale values ​​in region T using equation (2), and denot it as m. T ,in Let K represent the pixel grayscale value of the i-th maximum value in region T, and let K be 3. The block mean of the corresponding blocks in regions H and B is calculated using equations (3)-(4). and Where (x,y) are the coordinates of the pixel within the sliding window, f(x,y) represents the pixel grayscale value at pixel (x,y), j represents the index of the local block in region H, k represents the index of the local block in region B, and l represents the side length of the local block, with l being 3.

[0012]

[0013]

[0014]

[0015] Step 4: Calculate the local contrast ID between region T and region H using equations (5)-(6). TH and the local contrast ID between region T and region B TB ;

[0016]

[0017]

[0018] Step 5: Calculate the block contrast I at (x,y) using formula (7). FLLCM (x,y), the resulting contrast saliency plot is notated as I FLLCM ;

[0019] I FLLCM (x,y)=IDTB ×0.3+ID TH ×0.7 (7)

[0020] Step 6: Apply a median filter with a window size of 3×3 to image I. FLLCM The resulting image after denoising is denoted as

[0021] Step 7: Apply a mean filter to the image with a window size of 3×3. The resulting image after denoising is denoted as

[0022] Step 8: Calculate the adaptive threshold Th using equation (8), where Representing an image The mean, Table Image standard deviation As a regulating factor, k FLLCM ∈[10,35]

[0023]

[0024] Step 9: Using images Locate small targets with Th and traverse the image. If each point in the middle, If the pixel is not identified as the target pixel, it is considered a background pixel.

[0025] The advantages of this invention are:

[0026] 1. Effectively improves performance when detecting large-sized irregular targets;

[0027] 2. It can effectively suppress various kinds of interference in the background;

[0028] 3. Good detection results can be obtained by utilizing the characteristics of each part of saliency without fusing other information. Attached Figure Description

[0029] Figure 1 This is a flowchart of the present invention;

[0030] Figure 2 The simulation results of this algorithm in different scenarios are shown. Detailed implementation method:

[0031] The invention will now be described in detail with reference to specific examples.

[0032] 1: Read the original infrared image I0. The size of the image is M×N, where M is the width of image I0 in pixels and N is the height of image I0 in pixels.

[0033] 2: Design a sliding window to scan the image I0. The sliding window consists of 49 local blocks arranged in 7×7. Each local block consists of 3×3 pixels. The local blocks in the sliding window are numbered as shown in matrix C in equation (1). The block with the number T is called the T region, the block with the number G is called the G region, the block with the number H is called the H region, and the block with the number B is called the B region.

[0034]

[0035] 3: Calculate the average value of the three largest pixel grayscale values ​​in region T using equation (2), and denot it as m. T ,in Let K represent the pixel grayscale value of the i-th maximum value in region T, and let K be 3. The block average of the corresponding blocks in regions H and B is calculated using equations (3)-(4). and Where (x,y) are the coordinates of the pixel within the sliding window, f(x,y) represents the pixel grayscale value at pixel (x,y), j represents the index of the local block in region H, k represents the index of the local block in region B, and l represents the side length of the local block, with l being 3.

[0036]

[0037]

[0038]

[0039] 4: The local contrast ID between region T and region H is calculated using equations (5)-(6). TH And the local contrast ID between region T and region B TB ;

[0040]

[0041]

[0042] 5: The block contrast I at (x,y) is calculated using formula (7). FLLCM (x,y), the resulting contrast saliency plot is notated as I FLLCM ;

[0043] I FLLCM (x,y)=ID TB ×0.3+ID TH ×0.7 (7)

[0044] 6: Apply a median filter with a window size of 3×3 to image I FLLCM The resulting image after denoising is denoted as

[0045] 7: Apply a mean filter with a window size of 3×3 to the image. The resulting image after denoising is denoted as

[0046] 8: The adaptive threshold Th is calculated using equation (8), where Representing an image The mean, Table Image standard deviation As a regulating factor, k FLLCM ∈[10,35]

[0047]

[0048] 9: Utilizing images Locate small targets with Th and traverse the image. If each point in the middle, If the pixel is not identified as the target pixel, it is considered a background pixel.

Claims

1. A method for detecting small infrared targets based on four-layer windows with local contrast, characterized in that... The following steps are required: Step 1: Read the original infrared image I0. The size of the image is M×N, where M is the width of image I0 in pixels and N is the height of image I0 in pixels. Step 2: Design a sliding window to scan the image I0. The sliding window consists of 49 local blocks arranged in 7×7. Each local block consists of 3×3 pixels. The local blocks in the sliding window are numbered as shown in matrix C in equation (1). The block with the number T is called the T region, the block with the number G is called the G region, the block with the number H is called the H region, and the block with the number B is called the B region. Step 3: Calculate the average value of the three largest pixel grayscale values ​​in region T using equation (2), and denot it as m. T ,in The pixel grayscale value representing the i-th maximum value in region T, where K is 3, is used to calculate the block average of the corresponding blocks in regions H and B using equations (3)-(4). and Where (x,y) are the coordinates of the pixel within the sliding window, f(x,y) represents the pixel grayscale value at pixel (x,y), j represents the index of the local block in region H, k represents the index of the local block in region B, and l represents the side length of the local block, with l being 3. Step 4: Calculate the local contrast ID between region T and region H using equations (5)-(6). TH and the local contrast ID between region T and region B TB ; Step 5: Calculate the block contrast I at (x,y) using formula (7). FLLCM (x,y), the resulting contrast saliency plot is notated as I FLLCM ; I FLLCM (x,y)=ID TB ×0.3+ID TH ×0.7 (7) Step 6: Apply a median filter with a window size of 3×3 to image I. FLLCM The resulting image after denoising is denoted as Step 7: Apply a mean filter to the image with a window size of 3×3. The resulting image after denoising is denoted as Step 8: Calculate the adaptive threshold Th using equation (8), where Representing an image The mean, Table Image standard deviation As a regulating factor, k FLLCM ∈[10,35] Step 9: Using images Locate small targets with Th and traverse the image. If each point in the middle, If the pixel is not identified as the target pixel, it is considered a background pixel.

Images