An Infrared Image Background Suppression Method for Target Detection

Abstract

The present invention discloses an infrared image background suppression method for target detection, which relates to the technical fields of image processing and target detection. Aiming at the problems of the existing methods that fail to make good use of the time-domain information and have a relatively high false alarm rate, the present invention randomly intercepts blocks of the frame to be detected, calculates the relative displacement vectors to adjacent frames within a certain range, and then uses the obtained relative displacement vectors to calculate the background relative displacement vectors from the adjacent frames to the frame to be detected, thereby calculating the gray-scale weighted mean of the matching points to obtain the gray-scale response map of the adjacent frames. Then, the difference is taken with the frame to be detected, the absolute value is taken, and normalization is performed, and finally the detection result of the target is obtained. The present invention can offset the influence brought by the relative movement of the background to the detector, greatly reduce the false alarms generated by the background movement, and can well highlight the target in the response map.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing and target detection, and relates to an infrared image background suppression method for target detection. Background Art

[0002] Infrared search and track (IRST) plays an important role in many applications such as space-based surveillance systems, early warning systems, and military guidance systems. Existing methods can be roughly divided into two categories: detection-before-track (DBT) and track-before-detection (TBD).

[0003] The DBT method filters real targets by combining inter-frame information based on the single-frame detection result; the TBD method uses inter-frame information to accumulate the signal energy of the target along the trajectory to detect the target. Among them, the DBT method often relies too much on the single-frame detection result and cannot make good use of the time-domain information; while the TBD method often requires some motion information priors and has poor robustness.

[0004] Therefore, there is an urgent need for an infrared small target detection method that can make full use of time-domain information and has strong robustness. Summary of the Invention

[0005] In order to solve the problems existing in the above-mentioned prior art, the present invention intends to provide an infrared image background suppression method for target detection, aiming to solve the problems that the existing methods fail to make good use of time-domain information and have a high false alarm rate.

[0006] An infrared image background suppression method for target detection, preferably, includes the following steps:

[0007] A. For the infrared small target image to be detected, except for the frame to be detected itself, read the 3l-th frame, 2l-th frame, and l-th frame counted forward, and the 3l-th frame, 2l-th frame, and l-th frame counted backward (L is preferably 1 to 3);

[0008] B. Randomly select several blocks in the frame to be detected, and calculate the relative displacement vectors of each block from the frame to be detected to the adjacent frames respectively;

[0009] C. For each adjacent frame of the frame to be detected, calculate the background relative displacement vector of the adjacent frame relative to the frame to be detected according to the block relative displacement vectors obtained in step B;

[0010] D. Traverse each pixel in the frame to be detected, and calculate the gray-scale weighted mean of the matching points of the adjacent frame according to the background relative displacement vectors obtained in step C to obtain the gray-scale response map of the adjacent frame;

[0011] E. The original image of the frame to be detected is subtracted from the grayscale response images of adjacent frames, and the absolute value is taken and normalized to obtain the detection result image of the target.

[0012] Further, the specific steps of step B are as follows:

[0013] Step 1: Read the size of a single-frame image, where the height is denoted as h and the width is denoted as w; the preset number of blocks is N (preferably N = 10), the size of the block is patch_size×patch_size, and the search range parameter is R (note: patch_size should be large enough to ignore the influence of the relative movement of the target with respect to the background; preferably patch_size≥50, R≥10);

[0014] Step 2: For the l-th frame forward and the l-th frame backward of the frame to be detected (hereinafter simply referred to as the ±l frames), respectively randomly intercept N blocks of size patch_size×patch_size within the range [R:h - R, R:w - R] of the frame to be detected, and denote the upper-left coordinates of each block as For each block, respectively within the range of the ±l frames, search for the matching block with the criterion of the minimum mean square error, and denote the upper-left coordinates of each matching block as Thus, the relative displacement vectors of each block within the ±l frames are obtained

[0015] Step 3: For the 2l-th frame forward and the 2l-th frame backward of the frame to be detected (hereinafter simply referred to as the ±2l frames), respectively randomly intercept N blocks of size patch_size×patch_size within the range [2R:h - 2R, 2R:w - 2R] of the frame to be detected, and denote the upper-left coordinates of each block as For each block, respectively within the range of the ±2l frames, search for the matching block with the criterion of the minimum mean square error, and denote the upper-left coordinates of each matching block as Thus, the relative displacement vectors of each block within the ±2l frames are obtained

[0016] Step 4: For the 3l-th frame forward and the 3l-th frame backward of the frame to be detected (hereinafter simply referred to as the ±3l frames), respectively randomly intercept N blocks of size patch_size×patch_size within the range [3R:h - 3R, 3R:w - 3R] of the frame to be detected, and denote the upper-left coordinates of each block as For each block, respectively within the range of the ±3l frames, search for the matching block with the criterion of the minimum mean square error, and denote the upper-left coordinates of each matching block as The relative displacement vectors of each block within ±3l frames are thus obtained.

[0017] Furthermore, the specific operations of step C are as follows:

[0018] For each adjacent frame of the frame to be detected, the mean value of the relative displacement vectors of all blocks within it is taken as the background relative displacement vector of this adjacent frame relative to the frame to be detected.

[0019] Furthermore, the specific operations of step D are as follows:

[0020] M(i,j,k) = 0.3×[F(i + a -l ,j + b -l ,k - 1)+F(i + a +l ,j + b +l ,k + 1)] + 0.5×[F(i + a -2l ,j + b -2l ,k - 2)+F(i + a +2l ,j + b +2l ,k + 2)] + 0.2×[F(i + a -3l ,j + b -3l ,k - 3)+F(i + a +3l ,j + b +3l ,k + 3)]

[0021] In the formula, the first digit in the three - digit index is the row index, the second digit is the column index, and the third digit is the frame index; F represents the read - in original image sequence; a ±γ represents the background row relative displacement of ±γ frames, b ±γ represents the background column relative displacement of ±γ frames (γ = l, 2l, 3l); M(:,:,k) represents the adjacent - frame gray - level response map of the k - th frame.

[0022] Preferably, in step E, the frame to be detected is subtracted from the adjacent - frame gray - level response map, the absolute value is taken, and normalization is performed to obtain the detection result map of the target. The specific operations are as follows:

[0023] S(:,:,k) = |F(:,:,k)-M(:,:,k)|

[0024] In the formula, F(:,:,k) represents the original image of the k - th frame; M(:,:,k) represents the adjacent - frame gray - level response map of the k - th frame; S(:,:,k) represents the detection result map of the target of the k - th frame.

[0025] The beneficial effects of the present invention include:

[0026] 1. Estimate the background relative displacement vector of adjacent frames through randomly intercepted blocks, thereby calculate the gray weighted mean of the matching points of adjacent frames, obtain the gray response map of adjacent frames, then subtract it from the frame to be detected, take the absolute value and normalize it, and finally obtain the detection result map of the target.

[0027] 2. In the present invention, steps B and C estimate the background relative displacement vector of adjacent frames relative to the frame to be detected based on the relative displacement vectors of several random blocks, which can offset the influence brought by the background moving relative to the detector and greatly reduce the false alarms originating from the background movement.

[0028] 3. In the present invention, step D obtains the gray response of adjacent frames by calculating the gray weighted mean of the matching points of adjacent frames. The specially set weights fully consider the possible motion characteristics of the target. After combining with step E, the target can be well highlighted in the result map. Description of the Drawings

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use 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 therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0030] Figure 1 is the flowchart for background elimination of infrared images for target detection.

[0031] Figure 2 is the infrared small target image to be detected in Embodiment 1 of the present invention.

[0032] Figure 3 is the image of the first 3 frames to the last 3 frames (a total of 7 frames) of the infrared small target image to be detected in Embodiment 1 of the present invention.

[0033] Figure 4 is the gray response map of adjacent frames in Embodiment 1 of the present invention.

[0034] Figure 5 is the target detection result map in Embodiment 1 of the present invention. Detailed Embodiments

[0035] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part rather than all of the embodiments of this application. Usually, the components of the embodiments of this application described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents the selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative efforts fall within the scope of protection of this application.

[0036] Embodiment 1

[0037] The following will Figure 1 refer to Figure 5 the accompanying drawings to give a detailed description of the specific embodiments of the present invention:

[0038] A. For the infrared small target image to be detected, as Figure 2 shown, the real target is marked with a white box. Except for itself, read the 3l-th frame, 2l-th frame, and l-th frame counted forward, and the 3l-th frame, 2l-th frame, and l-th frame counted backward (l takes 1), as Figure 3 shown;

[0039] B. Randomly select several blocks in the frame to be detected, and calculate the relative displacement vectors of each block from the frame to be detected to the adjacent frames, a total of 6 adjacent frames;

[0040] Step 1: Read the size of a single-frame image h = 256, w = 256; preset the number of blocks N = 10, the size of the block is 50×50, and the search range parameter R = 10.

[0041] Step 2: For the first frame counted forward and the first frame counted backward (hereinafter simply referred to as ±1 frame) of the frame to be detected, randomly intercept 10 blocks with a size of 50×50 in the range of [10:246, 10:246] of the frame to be detected, and record the upper left coordinates of each block as For each block, respectively in the range of ±1 frame, search for the matching block with the minimum mean square error as the criterion, and record the upper left coordinates of each matching block as Thus, the relative displacement vectors of each block within the ±1 frame are obtained

[0042] Step 3: For the second frame counted forward and the second frame counted backward (hereinafter simply referred to as ±2 frame) of the frame to be detected, randomly intercept 10 blocks with a size of 50×50 in the range of [20:236, 20:236] of the frame to be detected, and record the upper left coordinates of each block as For each block, the In the range, the matching block is searched based on the minimum mean square error as the criterion, and the coordinates of the upper left corner of each matching block are recorded as Thus, the relative displacement vector of each block within ±2 frames is obtained.

[0043] Step 4: For the third frame forward and the third frame backward of the frame to be detected (hereinafter referred to as ±3 frames), randomly select 10 blocks of size 50×50 in the range [30:226,30:226] of the frame to be detected, and record the coordinates of the upper left corner of each block as For each block, the In the range, the matching block is searched based on the minimum mean square error as the criterion, and the coordinates of the upper left corner of each matching block are recorded as Thus, the relative displacement vector of each block within ±3 frames is obtained.

[0044] C. For each adjacent frame of the frame to be detected, calculate the background relative displacement vector of the adjacent frame relative to the frame to be detected according to the block relative displacement vector obtained in step B;

[0045] For each neighboring frame of the frame to be detected, the average of the relative displacement vectors of all blocks in the neighboring frame is taken as the background relative displacement vector of the neighboring frame relative to the frame to be detected.

[0046] D. Traverse each pixel in the frame to be detected, and calculate the grayscale weighted mean of the matching points in the adjacent frames according to the background relative displacement vector obtained in step C to obtain the grayscale response map of the adjacent frames, such as Figure 4 As shown;

[0047] M(i,j,k)=0.3×[F(i+a -1 ,j+b -1 ,k-1)+F(i+a +1 ,j+b +1 ,k+1)]+0.5×[F(i+a -2 ,j+b -2 ,k-2)+F(i+a +2 ,j+b +2 ,k+2)]+0.2×[F(i+a -3 ,j+b -3 ,k-3)+F(i+a +3 ,j+b +3 ,k+3)]

[0048] In the formula, the first bit of the three-digit index is the row index, the second bit is the column index, and the third bit is the frame index; F represents the original image sequence read in; a ±l represents the relative displacement of the background lines of ±l frames, b ±lThe relative displacement of the background columns representing ±l frames (l = 1, 2, 3); M(:, :, k) represents the gray response map of adjacent frames of the k-th frame.

[0049] E. The difference between the frame to be detected and the gray response map of adjacent frames is taken and the absolute value is normalized to obtain the detection result map of the target, as Figure 5 shown;

[0050] S(:, :, k) = |F(:, :, k) - M(:, :, k)|

[0051] In the formula, F(:, :, k) represents the original image of the k-th frame; M(:, :, k) represents the gray response map of adjacent frames of the k-th frame; S(:, :, k) represents the detection result map of the target of the k-th frame.

[0052] The above-described embodiments only represent the specific implementation manners of the present application, and the description thereof is relatively specific and detailed, but it should not be construed as a limitation on the protection scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the technical solution of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application.

Claims

1. An infrared image background suppression method for target detection, characterized in that Including the following steps: A. For the infrared image to be detected, except for the frame to be detected itself, read a finite number of frames forward and backward; B. Randomly select several blocks in the frame to be detected, and calculate the relative displacement vectors of each block from the frame to be detected to the adjacent frames respectively; C. For each adjacent frame of the frame to be detected, according to the relative displacement vectors of the blocks obtained in step B, calculate the background relative displacement vector of this adjacent frame relative to the frame to be detected; D. Traverse each pixel in the frame to be detected, and according to the background relative displacement vector obtained in step C, calculate the gray-scale weighted mean value of the matching points in the adjacent frame to obtain the gray-scale response map of the adjacent frame; E. Subtract the original image of the frame to be detected from the gray-scale response map of the adjacent frame, take the absolute value, and normalize it to obtain the detection result of the target; In step B, randomly select several blocks in the frame to be detected, and calculate the relative displacement vectors of each block from the frame to be detected to 6 adjacent frames respectively, which can be carried out through the following steps: Step 1: Read the size of a single-frame image, record the height as h and the width as w; preset the number of blocks as N, the size of the block as patch_size×patch_size, and the search range parameter as R; Step 2: For the l-th frame counted forward and the l-th frame counted backward with respect to the frame to be detected, hereinafter simply referred to as the ±l frames, randomly intercept N blocks of size patch_size×patch_size within the range of [R:h-R, R:w-R] of the frame to be detected, and denote the upper left coordinates of each block as For each block, respectively within the range of the ±l frames, search for the matching blocks with the criterion of the minimum mean square error, and denote the upper left coordinates of each matching block as Thus, the relative displacement vectors of each block within the ±l frames are obtained Step 3: For the 2l-th frame counted forward and the 2l-th frame counted backward from the frame to be detected, hereinafter simply referred to as the ±2l frames, randomly intercept N blocks of size patch_size×patch_size within the range of [2R:h - 2R, 2R:w - 2R] of the frame to be detected, and denote the upper left coordinates of each block as For each block, respectively within the range of the ±2l frames, search for the matching blocks with the criterion of minimizing the mean square error, and denote the upper left coordinates of each matching block as Thus, the relative displacement vectors of each block within the ±2l frames are obtained Step 4: For the 3l-th frame counted forward and the 3l-th frame counted backward from the frame to be detected, hereinafter simply referred to as the ±3l frames, randomly intercept N blocks of size patch_size×patch_size within the range of [3R:h - 3R, 3R:w - 3R] of the frame to be detected, and denote the upper left coordinates of each block as For each block, respectively within the range of the ±3l frames, search for the matching blocks with the criterion of the minimum mean square error, and denote the upper left coordinates of each matching block as Thus, the relative displacement vectors of each block within the ±3l frames are obtained 2. The infrared image background suppression method for target detection according to claim 1, wherein, In step A, for the infrared image to be detected, except for itself, the adjacent frame reading range is the 3l-th frame, the 2l-th frame, and the l-th frame counted forward from itself, and the 3l-th frame, the 2l-th frame, and the l-th frame counted backward from itself, making full use of the motion continuity information of the target between frames, where l is an integer.

3. The infrared image background suppression method for target detection according to claim 1, wherein In step C, for each adjacent frame of the frame to be detected, according to the relative displacement vectors of the blocks obtained in step B, calculate the background relative displacement vector of this adjacent frame relative to the frame to be detected. The specific operation is as follows: For each adjacent frame of the frame to be detected, take the mean value of the relative displacement vectors of all the blocks in it as the background relative displacement vector of this adjacent frame relative to the frame to be detected.

4. The infrared image background suppression method for target detection according to claim 1, wherein, In step D, traverse each pixel in the frame to be detected, and according to the background relative displacement vector obtained in step C, calculate the gray-scale weighted mean value of the matching points in the adjacent frame to obtain the gray-scale response map of the adjacent frame. The specific operation is as follows: M(i,j,k) = 0.3×[F(i + a -l , j + b -l , k - 1) + F(i + a +l , j + b +l , k + 1)] + 0.5×[F(i + a -2l , j + b -2l , k - 2) + F(i + a +2l , j + b +2l , k + 2)] + 0.2×[F(i + a -3l , j + b -3l , k - 3) + F(i + a +3l , j + b +3l , k + 3)] In the formula, i in the three - dimensional index is the row index, j is the column index, and k is the frame index; F represents the original image sequence read in; a ±γ represents the relative displacement of the background row of ±γ frames, b ±γ represents the relative displacement of the background column of ±γ frames, where γ = l, 2l, 3l; M(:, :, k) represents the gray - level response map of adjacent frames of the k - th frame.

5. The infrared image background suppression method for target detection according to claim 1, characterized in that In step E, subtract the frame to be detected from the gray-scale response map of the adjacent frame, take the absolute value, and normalize it to obtain the detection result of the target. The specific operation is as follows: S(:,:,k) = |F(:,:,k) - M(:,:,k)| In the formula, F(:,:,k) represents the original image of the k-th frame; M(:,:,k) represents the gray-scale response map of the adjacent frame of the k-th frame; S(:,:,k) represents the target detection result map of the k-th frame.

Images