Infrared small target detection method based on improved weighted enhanced local contrast measurement

Abstract

The invention discloses an infrared small target detection method based on improved weighted enhanced local contrast measurement, which is specifically as follows: step 1, using a nested window to calculate the final ELCM of each pixel in the original infrared image; step 2, weighted calculation IW is performed on the ELCM obtained in step 1; step 3, SIWELCM of the pixel (x, y) at each scale is calculated, and then the final IWELCM is obtained by maximum pooling operation; step 4, the infrared image processed by step 3 is normalized to a range of 0 to 255 gray levels; step 5, the threshold of the infrared image processed by step 4 is calculated by using an adaptive threshold segmentation method. The method of the invention solves the problem that false alarms are prone to occur in the prior art when detecting targets under complex backgrounds.

Description

Technical Field

[0001] The invention belongs to the technical field of infrared small target detection based on a single frame, and relates to an infrared small target detection method based on improved weighted enhanced local contrast measurement. Background Art

[0002] Infrared small target detection refers to obtaining the position of infrared small targets in an image through detection algorithms. With the development of computer vision and infrared imaging technology, infrared search and tracking (IRST) systems have been widely used in guidance, early warning, and traffic safety. Infrared small target detection plays a vital role in the performance of infrared search and tracking (IRST) systems.

[0003] In scenes with complex backgrounds, the long imaging distance, small target imaging area, lack of shape and texture information, and serious interferences such as bright background, background edge, and PNHB in complex scenes make this task very challenging.

[0004] In recent years, researchers have found that bionic methods have attracted widespread attention in existing single-frame-based detection algorithms. Biologists claim that the most important part of the human visual system is contrast rather than brightness. Bionic algorithms use contrast mechanisms to design effective algorithms to detect targets in infrared images. Summary of the invention

[0005] The purpose of the present invention is to provide an infrared small target detection method based on improved weighted enhanced local contrast measurement, which solves the problem of false alarms in the prior art when detecting targets under complex backgrounds.

[0006] The technical solution adopted by the present invention is an infrared small target detection method based on improved weighted enhanced local contrast measurement, which is specifically implemented according to the following steps:

[0007] Step 1, use a nested window to calculate the final ELCM of each pixel in the original infrared image;

[0008] Step 2, weighted calculation IW is performed on the ELCM obtained in step 1;

[0009] Step 3: Calculate the SIWELCM of the pixel (x, y) at each scale, and then obtain the final IWELCM through the maximum pooling operation;

[0010] Step 4, normalizing the infrared image processed in step 3 to a grayscale range of 0 to 255;

[0011] Step 5: Use the adaptive threshold segmentation method to calculate the threshold of the infrared image processed in step 4.

[0012] The present invention is also characterized in that

[0013] The specific process of step 1 is:

[0014] Step 1.1, set up a nested window, the nested window includes the target block T, and eight directional background blocks bi (i = 1, 2, ..., 8) are evenly set around the target block T. Then, the nested window is used to calculate the SELCM of each pixel (x, y). The SELCM of the current pixel (x, y) is defined as:

[0015]

[0016] In formula (1), M T Represents the average gray value of the K1 largest pixels in the target block T; mean bi Std represents the average gray value of all pixels in the background block bi in eight directions; b Represents the average value of the standard deviation between the eight directional background blocks bi; Represents a priori parameter with a value of 0.0001 to avoid division by 0; represents the enhancement coefficient of the target in the i-th direction in the nested window; (M T -mean bi ) 3 Used to amplify the difference between the target and the surrounding background block bi while ensuring the consistency of the symbol; Std b Used to improve the signal-to-noise ratio of the image; the minimum operation between eight directions is used to suppress background edges;

[0017] in,

[0018]

[0019] In formula (2), K1 represents the maximum number of gray levels of the target block T, is the qth maximum grayscale of the target block T;

[0020]

[0021] In formula (3), N b is the number of pixels contained in the eight-directional background block bi; It is the qth maximum gray value of the background block bi in eight directions;

[0022]

[0023] In formula (4), Std bi It is the average value of the grayscale standard deviation between the eight directional background blocks bi;

[0024]

[0025] Step 1.2, a non-negative constraint is used to obtain the final ELCM of each pixel (x, y) in the original infrared image, which is expressed as:

[0026] ELCM(x, y)=max(0, SELCM(x, y)) (6).

[0027] The specific process of step 2 is:

[0028] Step 2.1, when weighting ELCM by considering the target feature WT, the variance of the target block T is used, and the WT of the pixel point (x, y) is defined as:

[0029]

[0030] In formula (7), f(c, d) represents the gray value at the pixel point (c, d); (x, y) represents the pixel point as the center pixel point of the target block T; mean T It represents the average grayscale of all pixels in the target block T; the size of the target block T is the same as that of the eight directional background blocks bi, which is n×n;

[0031] Step 2.2, when considering the statistical difference WTB between the target and the surrounding background, the block grayscale ratio is used to obtain an improved weighting function IW;

[0032] For the center block T and eight directional background blocks bi, the block grayscale ratio is defined as:

[0033]

[0034]

[0035] In formula (9), M bi represents the average grayscale of K1 maximum pixels in the eight-directional background block bi, in Represents the qth maximum grayscale of the background block bi in eight directions;

[0036] The statistical difference of a pixel (x, y) is defined as:

[0037] WTB(x,y)=max{0,min{(BGC T -BGC bi )×(mean T -mean bi )}}, i = 1, 2, ..., 8 (10)

[0038] The improved weighting function IW of pixel point (x, y) is defined as:

[0039] IW(x,y)=WT(x,y)×WTB(x,y) (11).

[0040] The specific process of step 3 is:

[0041] Step 3.1, SIWELCM at a single scale of pixel point (x, y) is defined as:

[0042] SIWELCM(x,y)=IW(x,y)×ELCM(x,y)) (12)

[0043] Step 3.2, calculate the SIWELCM of the pixel (x, y) at each scale, and then obtain the final IWELCM through the maximum pooling operation, the expression is:

[0044] IWELCM(x,y)=max(SIWELCM p (x,y)) (13)

[0045] In formula (13), p represents the pth scale.

[0046] In step 4, the normalized expression is:

[0047]

[0048] In formula (14), max(I IWELCM ) is I IWELCM The maximum value of min(I IWELCM ) is I IWELCM The minimum value of .

[0049] The expression of the threshold in step 5 is:

[0050] Th=αmax(I IWELCM )+(1-α)mean(I IWELCM ) (15)

[0051] In formula (15), mean(I IWELCM ) is I IWELCM , the average value, α is a factor between 0 and 1, for single target detection, take 0.4≤α≤0.8;

[0052] When the I obtained by normalization in step 4 is IWELCM When it is not less than the threshold, it is an infrared image I IWELCM The goal is to normalize the I obtained in step 4. IWELCM When it is less than the set threshold, it is an infrared image I IWELCM background.

[0053] The beneficial effect of the present invention is that, aiming at the problem of infrared dim small target detection in complex background, the present invention proposes a new infrared small target detection framework, namely, improved weighted enhanced local contrast measurement; firstly, the enhanced local contrast measurement introduces the standard deviation of the surrounding background on the basis of the ratio difference combined with the local contrast measurement algorithm to improve the signal-to-clutter ratio (SCR) of the image; secondly, the weighting function weights the local contrast by the characteristics of the target and the statistical difference between the target and the surrounding background; finally, a simple adaptive threshold segmentation is used to extract the real target; experimental results on a real image data set show that the method of the present invention improves the signal-to-clutter ratio gain (SCRG) and background suppression (BSF) performance indicators, and also has a good performance on the ROC curve. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of the infrared small target detection method based on improved weighted enhanced local contrast measurement of the present invention;

[0055] Figure 2 It is a nested window used in the infrared small target detection method based on improved weighted enhanced local contrast measurement of the present invention;

[0056] Figure 3 It is the original infrared image in the infrared small target detection method based on improved weighted enhanced local contrast measurement of the present invention;

[0057] Figure 4 The processing results of each stage of the proposed method of the infrared small target detection method based on improved weighted enhanced local contrast measurement of the present invention;

[0058] Figure 5 The detection results of the infrared small target detection method based on improved weighted enhanced local contrast measurement of the present invention and the comparison method used in different data sets; wherein (a) represents the detection result of the MPCM method, (b) represents the detection result of the RLCM method, (c) represents the detection result of the WLCM method, (d) represents the detection result of the WSLCM method, and (e) represents the detection result of the method of the present invention;

[0059] Figure 6 It is the ROC curve of the infrared small target detection method based on the improved weighted enhanced local contrast measurement of the present invention and the adopted comparison method tested on the data set 1;

[0060] Figure 7 It is the ROC curve of the infrared small target detection method based on the improved weighted enhanced local contrast measurement of the present invention and the adopted comparison method tested on the data set 2;

[0061] Figure 8It is the ROC curve of the infrared small target detection method based on the improved weighted enhanced local contrast measurement of the present invention and the adopted comparison method tested on the data set three;

[0062] Fig. 9 This is the ROC curve of the infrared small target detection method based on improved weighted enhanced local contrast measurement of the present invention and the adopted comparison method tested on data set four. DETAILED DESCRIPTION

[0063] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0064] The present invention provides an infrared small target detection method based on improved weighted enhanced local contrast measurement, such as Figure 1 As shown, the specific implementation steps are as follows:

[0065] Step 1, use a nested window to calculate the original infrared image ( Figure 3 The final ELCM (Enhanced Local Contrast Measurement) for each pixel in the image (as shown);

[0066] Step 1.1, set up a nested window, such as Figure 2 As shown, the nested window includes the target block T, and eight directional background blocks bi (i=1, 2, ..., 8) are evenly arranged around the target block T. The nested window is used to calculate the SELCM (enhanced local contrast at a single scale) of each pixel (x, y). The SELCM of the current pixel (x, y) is defined as:

[0067]

[0068] In formula (1), M T Represents the average gray value of the K1 largest pixels in the target block T; mean bi Std represents the average gray value of all pixels in the background block bi in eight directions; b Represents the average value of the standard deviation between the eight directional background blocks bi; Represents a priori parameter with a value of 0.0001 to avoid division by 0; represents the enhancement coefficient of the target in the i-th direction in the nested window; (M T -mean bi ) 3 Used to amplify the difference between the target and the surrounding background block bi while ensuring the consistency of the symbol; Std b Used to improve the signal-to-noise ratio (SCR) of the image; the minimum operation between eight directions is used to suppress background edges;

[0069] in,

[0070]

[0071] In formula (2), K1 represents the maximum number of gray levels of the target block T, is the qth maximum grayscale of the target block T;

[0072]

[0073] In formula (3), N b is the number of pixels contained in the eight-directional background block bi; It is the qth maximum gray value of the background block bi in eight directions;

[0074]

[0075] In formula (4), Std bi It is the average value of the grayscale standard deviation between the eight directional background blocks bi;

[0076]

[0077] Step 1.2, a non-negative constraint is used to obtain the final ELCM of each pixel (x, y) in the original infrared image, which is expressed as:

[0078] ELCM(x, y)=max(0, SELCM(x, y)) (6)

[0079] Step 2, performing weighted calculation IW (improved weighting function) on the ELCM obtained in step 1;

[0080] Generally speaking, the contrast of small targets in the original infrared image is always very low, and the detection result of ELCM alone always has many false alarm points. Therefore, the present invention proposes an improved weighting function framework to weight ELCM, so that the proposed algorithm can achieve an ideal effect. For the design of the weighting function, two aspects are considered: (1) the characteristics of the target; (2) the statistical difference between the target and the surrounding background; so as to better enhance the target and suppress the background;

[0081] Step 2.1, when weighting ELCM considering the target feature (WT), the variance of the target block T is used, and the WT of the pixel point (x, y) is defined as:

[0082]

[0083] In formula (7), f(c, d) represents the gray value at the pixel point (c, d); (x, y) represents the pixel point as the center pixel point of the target block T; mean T It represents the average grayscale of all pixels in the target block T; the size of the target block T is the same as that of the eight directional background blocks bi, which is n×n;

[0084] Step 2.2, when considering the statistical difference between the target and the surrounding background (WTB), the block grayscale ratio (BGC) is used to obtain an improved weighting function IW;

[0085] For the center block T and eight directional background blocks bi, BGC is defined as:

[0086]

[0087]

[0088] In formula (9), M bi represents the average grayscale of K1 maximum pixels in the eight-directional background block bi, in Represents the qth maximum grayscale of the background block bi in eight directions;

[0089] The WTB of a pixel (x, y) is defined as:

[0090] WTB(x, y) = max{0, min{(BGC T -BGC bi )×(mean T -mean bi ))}, i = 1, 2, ..., 8 (10)

[0091] The improved weighting function IW of pixel point (x, y) is defined as:

[0092] IW(x,y)=WT(x,y)×WTB(x,y) (11)

[0093] Step 3: Calculate the weighted local contrast (SIWELCM) of the pixel (x, y) at each scale, and then obtain the final IWELCM through the maximum pooling operation;

[0094] Step 3.1, SIWELCM at a single scale of pixel point (x, y) is defined as:

[0095] SIWELCM(x,y)=IW(x,y)×ELCM(x,y)) (12)

[0096] Step 3.2, in practical applications, the size of small targets in infrared images is usually unknown, so multi-scale operations must be performed. The so-called multi-scale targets here refer to the size range that can be taken in infrared images;

[0097] Calculate the SIWELCM of the pixel (x, y) at each scale, and then obtain the final IWELCM through the maximum pooling operation, which is expressed as:

[0098] IWELCM(x,y)=max(SIWELCM p (x,y)) (13)

[0099] In formula (13), p represents the pth scale;

[0100] Step 4: The infrared image (I IWELCM ) is normalized to the range of 0 to 255 gray levels, and the expression is:

[0101]

[0102] In formula (14), max(I IWELCM ) is I IWELCM The maximum value of min(I IWELCM ) is I IWELCM The minimum value of

[0103] I IWELCM (x, y) is the same as IWELCM(x, y), which represents the value at the pixel point (x, y) in the infrared image; I IWELCM represents the entire infrared image obtained after processing in step 3;

[0104] Step 5, using the adaptive threshold segmentation method to calculate the threshold of the infrared image processed in step 4, when the I IWELCM When it is not less than the threshold, it is an infrared image I IWELCM The goal is to normalize the I obtained in step 4. IWELCM When it is less than the set threshold, it is an infrared image I IWELCM The background of the threshold is:

[0105] Th=αmax(I IWELCM )+(1-α)mean(I IWELCM ) (15)

[0106] In formula (15), mean(I IWELCM ) is I IWELCM The average value of , α is a factor between 0 and 1, and 0.4≤α≤0.8 is appropriate for single target detection;

[0107] For each infrared image I IWELCM Saliency map, the larger the value, the more likely it is the target.

[0108] like Figure 4 As shown, from left to right are the original infrared image, the results after step 3 and step 5 processing, it can be seen that the detection rate of the present invention is high and the false alarm rate is low;

[0109] like Figure 6-Figure 9 As shown in the figure, they are ROC curves tested on different data sets respectively; it can be seen from the figure that the method adopted by the present invention can achieve high detection rate and low false alarm rate under different data sets, and the detection performance ranks among the top among the several comparative methods adopted;

[0110] The present invention is based on an infrared small target detection method based on improved weighted enhanced local contrast measurement, which can effectively extract real infrared weak small targets from interference objects and has better detection performance. First, by combining the local contrast mechanism with the calculation of the signal-to-clutter ratio (SCR), an enhanced local contrast measurement method is proposed, which can enhance the suspected infrared weak small target area in the image while also improving the signal-to-clutter ratio (SCR) of the image. Secondly, by utilizing the characteristics of weak small targets in infrared images and the statistical differences between the target and the surrounding background, an improved weighting function is proposed to further enhance the target and suppress the background. Finally, an adaptive threshold segmentation method is used to obtain the detected target. Comparative experiments on data sets of different scenes show that the method adopted by the present invention is compared with four existing popular methods (MPCM, RLCM, WLCM, WSLCM), such as Figure 5 As shown, the method of the present invention has better detection performance.

Claims

1. An infrared small target detection method based on improved weighted enhanced local contrast measurement, characterized in that: Follow the steps below to implement it: Step 1, use a nested window to calculate the final ELCM of each pixel in the original infrared image; The specific process of step 1 is: Step 1.1, set up a nested window, the nested window includes the target block T, and eight directional background blocks bi are evenly set around the target block T, i=1, 2, ..., 8, then the nested window is used to calculate the SELCM of each pixel (x, y), and the SELCM of the current pixel (x, y) is defined as: （1） In formula (1), Represents the average gray value of the K1 largest pixels in the target block T; Represents the average gray value of all pixels in the background block bi in eight directions; Represents the average value of the standard deviation between the eight directional background blocks bi; Represents a priori parameter with a value of 0.0001 to avoid division by 0; Represents the enhancement coefficient of the target in the i-th direction in the nested window; Used to amplify the difference between the target and the surrounding background block bi while ensuring the consistency of the symbol; Used to improve the signal-to-noise ratio of the image; the minimum operation between eight directions is used to suppress background edges; in, （2） In formula (2), K1 represents the maximum number of gray levels of the target block T. is the qth maximum grayscale of the target block T; （3） In formula (3), is the number of pixels contained in the eight-directional background block bi; It is the qth maximum gray value of the background block bi in eight directions; （4） In formula (4), It is the average value of the grayscale standard deviation between the eight directional background blocks bi; （5） Step 1.2, a non-negative constraint is used to obtain the final ELCM of each pixel (x, y) in the original infrared image, which is expressed as: （6）； Step 2, weighted calculation IW is performed on the ELCM obtained in step 1; The specific process of step 2 is: Step 2.1, when weighting ELCM by considering the target feature WT, the variance of the target block T is used, and the WT of the pixel point (x, y) is defined as: （7） In formula (7), represents the gray value at the pixel point (c, d); (x, y) represents the pixel point as the center pixel point of the target block T; It represents the average grayscale of all pixels in the target block T; the size of the target block T is the same as that of the eight directional background blocks bi, which is ; Step 2.2, when considering the statistical difference WTB between the target and the surrounding background, the block grayscale ratio is used to obtain an improved weighting function IW; For the center block T and eight directional background blocks bi, the block grayscale ratio is defined as: （8） （9） In formula (9), represents the average grayscale of K1 maximum pixels in the eight-directional background block bi, ,in Represents the qth maximum grayscale of the background block bi in eight directions; The statistical difference of a pixel (x, y) is defined as: （10） The improved weighting function IW of pixel point (x, y) is defined as: （11）； Step 3: Calculate the SIWELCM of the pixel (x, y) at each scale, and then obtain the final IWELCM through the maximum pooling operation; Step 4, normalizing the infrared image processed in step 3 to a grayscale range of 0 to 255; Step 5: Use the adaptive threshold segmentation method to calculate the threshold of the infrared image processed in step 4.

2. The infrared small target detection method based on improved weighted enhanced local contrast measurement according to claim 1 is characterized in that: The specific process of step 3 is: Step 3.1, SIWELCM at a single scale of pixel point (x, y) is defined as: （12） Step 3.2, calculate the SIWELCM of the pixel (x, y) at each scale, and then obtain the final IWELCM through the maximum pooling operation, the expression is: （13） In formula (13), p represents the pth scale.

3. The infrared small target detection method based on improved weighted enhanced local contrast measurement according to claim 1 is characterized in that: In step 4, the normalized expression is: （14） In formula (14), for The maximum value of for The minimum value of .

4. The infrared small target detection method based on improved weighted enhanced local contrast measurement according to claim 1 is characterized in that: The expression of the threshold in step 5 is: （15） In formula (15), for The average value of is a factor between 0 and 1, and for single target detection, it is 0.4≤ ≤0.8; When the normalization process in step 4 is obtained If it is not less than the threshold, it is an infrared image. The goal is to get When it is less than the set threshold, it is an infrared image. background.

Images