Infrared small target detection algorithm based on multi-scale differential contrast enhancement

Abstract

The invention discloses an infrared small target detection algorithm based on multi-scale differential contrast enhancement, which belongs to the technical field of image segmentation and comprises the following steps: acquiring an original image containing a single-frame infrared small target; performing multi-scale Gaussian blur processing on the original image to generate a series of blurred images with different scales; calculating the difference of the adjacent scale blurred images, and generating a multi-scale Gaussian difference image; detecting a local maximum value point in the multi-scale Gaussian difference image; constructing a multi-scale block region for each local maximum value point, and calculating the mean difference between the local maximum value point and eight neighborhoods thereof by using a multi-scale block contrast function; summarizing calculation results of the multi-scale block contrast function to generate a contrast enhanced image; and setting a threshold value according to the average value and the standard deviation of the contrast-enhanced image, and segmenting the contrast-enhanced image according to the threshold value to obtain a final target detection result. The method can effectively detect the infrared small target in the complex background, and has good robustness and accuracy.

Description

Technical Field

[0001] The invention relates to the technical fields of image processing, target detection and image segmentation, and in particular to an infrared small target detection algorithm based on multi-scale differential contrast enhancement. Background Art

[0002] Multiscale Difference of Gaussian (DoG) is a feature detection algorithm widely used in image processing and computer vision. DoG highlights the salient features in the image by Gaussian blurring the image at different scales and then calculating the difference between blurred images at adjacent scales. Multiscale DoG has multiscale characteristics and can effectively detect and capture small target areas in the image. In infrared small target detection, multiscale DoG can adapt to small targets of different sizes and contrasts and enhance their saliency in complex backgrounds. However, the use of the DoG method alone may be interfered by background noise and non-target areas, thus affecting the detection effect.

[0003] Multiscale Patch-based Contrast Measure (MPCM), as a small target detection method based on local contrast enhancement, enhances local contrast by calculating the grayscale mean difference between the target area and its eight neighborhoods. MPCM constructs block areas at multiple scales and performs contrast calculations on these block areas, which can more accurately locate and enhance small targets in the image. In infrared small target detection, MPCM combines multi-scale information to enhance the saliency of the target at different scales, thereby improving detection accuracy. The inventors of the present application have found that by integrating multi-scale DoG and MPCM methods, small targets in infrared images can be effectively highlighted under complex backgrounds, achieving more robust target detection. Summary of the invention

[0004] The purpose of the present invention is to provide an infrared small target detection algorithm based on multi-scale differential contrast enhancement, which can suppress background clutter and improve the target detection rate in the infrared small target detection under complex background.

[0005] The technical solution adopted by the present invention is: an infrared small target detection algorithm based on multi-scale differential contrast enhancement, comprising the following steps:

[0006] Step 1: Obtain an original image containing a single frame of a small infrared target;

[0007] Step 2: Perform multi-scale Gaussian blur processing on the original image to generate a series of blurred images of different scales;

[0008] Step 3: Calculate the difference of adjacent scale blurred images to generate a multi-scale Gaussian difference image;

[0009] Step 4: Detect local maximum points in the multi-scale Gaussian difference image to identify significant feature areas;

[0010] Step 5: For each local maximum point, construct a multi-scale block region, and use the multi-scale block contrast function to calculate the mean difference between the local maximum point and its eight neighborhoods;

[0011] Step 6: Summarize the calculation results of the multi-scale block contrast function to generate a contrast enhanced image;

[0012] Step 7: Set a threshold according to the mean value and standard deviation of the contrast-enhanced image, segment the contrast-enhanced image according to the threshold, and obtain the final target detection result after segmentation.

[0013] The beneficial effects of the present invention compared with the prior art are:

[0014] (1) The present invention uses a multi-scale Gaussian difference (DoG) method to generate a multi-scale feature image, and applies a multi-scale block contrast function on the feature image to enhance the local contrast. This combination effectively utilizes multi-scale information and can better highlight small targets in infrared images.

[0015] (2) The present invention adaptively sets the binarization threshold by calculating the mean value and standard deviation of the contrast-enhanced image. This adaptive binarization method based on image statistical characteristics is more flexible than the fixed threshold method and can better adapt to the grayscale distribution of different images, thereby improving the accuracy of target detection.

[0016] (3) The present invention generates a contrast-enhanced image by calculating and integrating local contrast information at multiple scales. This method makes full use of multi-scale context information and effectively enhances the contrast of salient feature areas in the image. In addition, through multi-scale context information, small targets of different sizes can be detected more accurately and the false alarm rate can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the infrared small target detection algorithm based on multi-scale differential contrast enhancement of the present invention;

[0018] Figure 2 The detection results of different algorithms, (a) is the original image, (b) is the result of LCM algorithm processing, (c) is the result of Top-Hat algorithm processing, (d) is the result of RPCA algorithm processing, (e) is the result of this method processing;

[0019] Figure 3 SCRG values ​​of different algorithms for each image, where (a) shows the SCRG values ​​of each algorithm for Figure 2(a) shows the detection results of the first row of images, and (b) shows the detection results of each algorithm. Figure 2 (a) shows the detection results of the second row of images, and (c) shows the detection results of each algorithm. Figure 2 (a) Detection results of the third row of images. DETAILED DESCRIPTION

[0020] The present invention is further described below with reference to the accompanying drawings and specific embodiments.

[0021] Figure 1 FIG. 1 is a flowchart of an infrared small target detection algorithm based on multi-scale differential contrast enhancement of the present invention. Figure 1 As shown, the specific process steps of the infrared small target detection algorithm based on multi-scale differential contrast enhancement described in the present invention are as follows:

[0022] Step 1: Acquire an original image containing a single frame of a small infrared target. In the embodiment of the present invention, the target area of ​​the small infrared target in the original image is not greater than 25 pixels.

[0023] Step 2: Perform multi-scale Gaussian blur processing on the original image to generate a series of blurred images of different scales.

[0024] The Gaussian blurring process can include:

[0025] Set the number of scales to 10, and for each scale , calculate the corresponding standard deviation :

[0026] ,

[0027] in, is the scale increment factor.

[0028] Generate a 2D Gaussian blur kernel using standard deviation :

[0029] ,

[0030] in, are the horizontal and vertical coordinates of the image.

[0031] For the original image Apply a Gaussian blur kernel to generate a blurred image of the corresponding scale :

[0032] ,

[0033] in, Represents a convolution operation.

[0034] Step 3: Calculate the difference of blurred images at adjacent scales to generate a multi-scale Difference of Gaussian (DoG) image.

[0035] By calculating the difference between blurred images at adjacent scales (i.e., Gaussian difference), the edges and details in the blurred image can be enhanced.

[0036] For each pair of blurred images at adjacent scales and , calculate their difference image :

[0037] ,

[0038] in, Represents pixel coordinates in an image.

[0039] Step 4: Detect local maximum points in the multi-scale DoG image to identify significant feature areas.

[0040] For each pixel position , in scale Next, determine whether it is a local maximum point:

[0041] and ,

[0042] in, .

[0043] Step 5: For each local maximum point, construct a multi-scale block region, and use the multi-scale block contrast function to calculate the mean difference between the local maximum point and its eight neighbors.

[0044] set up The index of the local maximum point is ,in The value range is 1 to . The coordinates of the local maximum point are recorded as , Indicates the corresponding scale. The multi-scale block area is usually a The square area with a side length of ,in is a constant factor. For each local maximum point, a multi-scale block region is defined according to the corresponding scale :

[0045] ,

[0046] Calculate the mean of multi-scale block regions and the mean of the eight neighborhoods :

[0047] ,

[0048] ,

[0049] in Indicates a Neighborhood (excluding the center point ).

[0050] Calculate contrast :

[0051] .

[0052] Step 6: Summarize the calculation results of the multi-scale block contrast function to generate a contrast enhanced image.

[0053] Assume the original image Size ,have local maximum points, indexed by ,in The value range is 1 to . Initialize a size of The all-zero matrix and a size of The all-zero matrix . Traverse each local maximum point and calculate the contrast value Add to The corresponding position . Traverse the matrix For each pixel in the third dimension, find the pixel point The maximum value in this dimension is assigned to the matrix The corresponding Position. Complete the traversal matrix To enhance the contrast of the image, the salient feature areas are clearly highlighted.

[0054] Step 7: Set a threshold according to the mean value and standard deviation of the contrast-enhanced image, segment the contrast-enhanced image according to the threshold, and obtain the final target detection result after segmentation.

[0055] Compute the mean of the contrast enhanced image and standard deviation . Choose a threshold based on the mean and standard deviation :

[0056] ,

[0057] in, is a control parameter that adjusts the sensitivity. Segmentation of contrast enhanced images :The gray value is less than The pixels of are set to zero, and the other pixel values ​​are retained, and finally the target detection result is obtained.

[0058] To prove the effectiveness and detection capability of this method, three infrared small target images in complex backgrounds were selected for verification and compared with other commonly used target detection methods. The other three comparison algorithms are: Local Contrast Measure (LCM), Top-Hat filtering and RPCA. Figure 2 As shown in the figure, (a) is the original image, (b) is the result of LCM algorithm processing, (c) is the result of Top-Hat algorithm processing, (d) is the result of RPCA algorithm processing, and (e) is the result of this method processing.

[0059] In order to quantitatively compare the advantages and disadvantages of various algorithms, the signal-to-clutter ratio (SCR) and signal-to-clutter ratio gain (SCRG) are introduced:

[0060] ,

[0061] ,

[0062] in, is the mean gray value of the target area, is the mean gray value of the background area, is the standard deviation of the grayscale value of the background area, Represents the clutter signal ratio of the original image, It represents the clutter signal ratio of the output image after algorithm processing. SCRG represents the enhancement effect of the algorithm on the target. The larger the SCRG value, the better the detection effect of the algorithm.

[0063] Figure 3 The SCRG values ​​in the figure show the effects of each algorithm on Figure 2 (a) shows the detection results of three original images, where (a) shows the detection results of each algorithm. Figure 2 (a) shows the detection results of the first row of images, and (b) shows the detection results of each algorithm. Figure 2 (a) shows the detection results of the second row of images, and (c) shows the detection results of each algorithm. Figure 2 (a) The detection results of the third row of images. Figure 2 and Figure 3 , which proves that this method has good detection capability for small infrared targets in complex backgrounds.

[0064] The above is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technology can understand and think of any changes or substitutions within the technical scope disclosed by the present invention, which should be included in the scope of the present invention. The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

Claims

1. An infrared small target detection algorithm based on multi-scale differential contrast enhancement, characterized in that: The following steps are involved: Step 1: Obtain an original image containing a single frame of a small infrared target; Step 2: Perform multi-scale Gaussian blur processing on the original image to generate a series of blurred images of different scales; Step 3: Calculate the difference of adjacent scale blurred images to generate a multi-scale Gaussian difference image; Step 4: Detect local maximum points in the multi-scale Gaussian difference image to identify significant feature areas; Step 5: For each local maximum point, construct a multi-scale block region, and use the multi-scale block contrast function to calculate the mean difference between the local maximum point and its eight neighborhoods; Step 6: Summarize the calculation results of the multi-scale block contrast function to generate a contrast enhanced image; Step 7: Set a threshold according to the mean value and standard deviation of the contrast-enhanced image, segment the contrast-enhanced image according to the threshold, and obtain the final target detection result after segmentation.

2. The infrared small target detection algorithm based on multi-scale differential contrast enhancement according to claim 1 is characterized in that: The target area of ​​the small infrared target in the original image is no larger than 25 pixels.

3. The infrared small target detection algorithm based on multi-scale differential contrast enhancement according to claim 1 is characterized in that: The Gaussian blurring process includes: Set the number of scales to 10, and for each scale i, calculate the corresponding standard deviation : ， in, is the scale increment factor; Generate a 2D Gaussian blur kernel using standard deviation : ， in, is the horizontal and vertical coordinates of the image; Apply a Gaussian blur kernel to the original image I to generate a blurred image of the corresponding scale : ， in, Represents a convolution operation.

4. The infrared small target detection algorithm based on multi-scale differential contrast enhancement according to claim 3 is characterized in that: For each pair of blurred images at adjacent scales and , calculate their multi-scale Gaussian difference images : ， in, Represents pixel coordinates in an image.

5. The infrared small target detection algorithm based on multi-scale differential contrast enhancement according to claim 4 is characterized in that: Step 4 includes: For each pixel position , determine whether it is a local maximum point at scale i: as well as , in, .

6. The infrared small target detection algorithm based on multi-scale differential contrast enhancement according to claim 5 is characterized in that: Step 5 includes: let the index of the r local maximum points be j, where the value range of j is 1 to r, and the corresponding scale is , the coordinates of the local maximum point are recorded as , the multi-scale block area is a The square area with a side length of ,in is a constant factor. For each local maximum point, a multi-scale block area is defined according to the corresponding scale. : ， Calculate the mean of multi-scale block regions and the mean of the eight neighborhoods : ， ， in, Represents a 3×3 neighborhood, excluding the center point , Using the multi-scale block contrast function to calculate the mean difference between the local maximum point and its eight neighbors includes: calculating the contrast : 。 7. The infrared small target detection algorithm based on multi-scale differential contrast enhancement according to claim 6 is characterized in that: Step six includes: assuming the original image Size ,exist local maximum points, indexed by ,in The value range is 1 to , initialize a size of The all-zero matrix and a size of The all-zero matrix , traverse each local maximum point, and calculate the contrast Add to The corresponding position , traverse the matrix For each pixel in the third dimension, find the pixel point The maximum value in this dimension is assigned to the matrix The corresponding Position, complete the traversal matrix Enhances images for contrast.

8. The infrared small target detection algorithm based on multi-scale differential contrast enhancement according to claim 7 is characterized in that: Step seven includes: Compute the mean of the contrast enhanced image and standard deviation , select a threshold based on the mean and standard deviation : ， in, is a control parameter used to adjust the sensitivity; Based on the threshold Segmentation of contrast enhanced images , and obtain the final target detection result.

Images