A threshold segmentation method for infrared small target background suppression post image

Abstract

The application discloses a threshold segmentation method for an infrared small target background suppressed image, which realizes threshold segmentation of the infrared small target image after background suppression by calculating the similarity between a Gaussian filter template and an image block, fully utilizes the gray scale features and structural features of the small target, and accurately extracts all the small targets in the image. The application improves the problem that the traditional threshold segmentation method is prone to missing small targets in a multi-target detection task, realizes image segmentation with high detection rate and high real-time rate, provides guarantee for accurate detection of the infrared small target, and has the advantages of simple segmentation process, high efficiency and the like.

Description

Technical Field

[0001] This invention relates to a threshold segmentation processing method for achieving high detection rates of small infrared targets after complex background suppression. Background Technology

[0002] With the application and development of infrared early warning systems, threshold segmentation, as a key technology in the infrared small target detection process, has been widely used in these systems. The general steps for infrared small target detection are to first suppress the background of the image to separate the target from the background, and then use threshold segmentation to further improve the image's contrast and signal-to-noise ratio. However, when there are multiple small targets in the image and the grayscale difference between the targets is too large after background suppression, traditional threshold segmentation methods may miss some targets. Summary of the Invention

[0003] To address the aforementioned issues, this application aims to propose a threshold segmentation method for infrared small target detection, thereby improving the problem of traditional threshold segmentation methods easily missing small targets in multi-target detection tasks, increasing the algorithm's detection rate, and providing assurance for infrared target detection, security surveillance, and other applications.

[0004] To achieve the above objectives, the threshold segmentation method of the present invention includes the following steps:

[0005] Step 1: Generate a Gaussian filter template P of size K×K. g and using P g The histogram distribution generates a one-dimensional vector V. h ;

[0006] Step 2: Normalize the input image I to obtain I p Using a sliding window of size K×K to traverse I p m image blocks P are obtained. i Let i = 1, 2, ..., m. Generate m one-dimensional vectors V based on the histograms of the m image patches respectively. i , i = 1, 2, ..., m.

[0007] Step 3: Calculate V using cosine similarity. h With V i Similarity S between i The image I is binarized to obtain the threshold segmentation map R at scale K. K .

[0008] Step 4: Calculate the threshold segmentation map R at four scales. K The final threshold segmentation result R is obtained by using max pooling.

[0009] The beneficial effects of this invention are:

[0010] (1) This invention utilizes the grayscale and structural features of the small target itself to segment the background suppression results, which effectively improves the problem of target omission in multi-target detection, increases the detection rate, and reduces the false alarm rate.

[0011] (2) This invention utilizes multi-scale operations to adapt to small targets of different sizes, effectively improving the reliability of the method. Moreover, the method is simple, efficient, and easy to implement. Attached Figure Description

[0012] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0013] Figure 1 This is a flowchart of the present invention;

[0014] Figure 2 The four scales of Gaussian filter templates are generated;

[0015] Figure 3 Background-suppressed infrared small target images for testing purposes;

[0016] Figure 4 This is a comparison chart of the threshold segmentation results. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and specific examples.

[0018] Step 1: Generate a Gaussian filter template P of size K×K. g Obtain the histogram G of the Gaussian filter template. h and histogram G h Divide into M regions, each region consisting of 256 / M consecutive gray levels;

[0019] Step 2: Histogram G of the Gaussian filter template h , for G h The gray values ​​in M ​​regions are summed to obtain M data points, which are then combined into a one-dimensional vector V. h ;

[0020] Step 3: For the input image I, such as... Figure 3 As shown, a normalization operation is performed, making the maximum pixel value of the image 255 and the minimum value 0, to obtain the preprocessed image I. p ;

[0021] Step 4: Traverse image I pixel by pixel using a sliding window of size K×K. p With a sliding step size of 1, m infrared image blocks P are obtained. i , i = 1, 2, ..., m; obtain each input infrared image block P respectivelyi Histogram H i Let i = 1, 2, ..., m, and then set the histogram H... i Divide into N regions, each region consisting of 256 / N consecutive gray levels;

[0022] Step 5: Histogram H for each image patch i , for H i The gray values ​​in N regions are summed to obtain N data points, which are then combined into a one-dimensional vector V. i , i = 1, 2, ..., m;

[0023] Step 6: Calculate V using the cosine similarity metric. h and V i Similarity between S i , i = 1, 2, ..., m;

[0024] Step 7: Compare the similarity S respectively. i The image is binarized using a manually set threshold T to obtain a thresholded segmentation map R at scale K. K ;

[0025] Step 8: Repeat the above seven steps for each of the four different values ​​of K to obtain four threshold segmentation results R. K , for R K Perform max pooling to obtain the final threshold segmentation result R.

[0026] In step one of this invention, M and N are both 4; K is 3, 5, 7, and 9 respectively, and the corresponding Gaussian filter template at each scale is... Figure 2 As shown, it is designed to accommodate small infrared targets of different sizes.

[0027] The specific calculation method for cosine similarity in step six of this invention is shown in the following formula:

[0028]

[0029] Among them, S i For the image block P corresponding to the i-th pixel point i With Gaussian filter template P g similarity, V i Let V be the histogram vector of the i-th image patch. h is the histogram vector of the Gaussian filter template, and m is the total number of pixels in the image.

[0030] The normalization operation in step three of this invention is specifically shown in the following formula:

[0031]

[0032] Among them, I p Let I be the normalized image, and let min{I} and max{I} be the minimum and maximum pixel values ​​of the input image, respectively.

[0033] The binarization operation in step six of this invention is specifically shown in the following formula:

[0034]

[0035] in, The result is the binarization of the i-th pixel at scale K. T is a manually set threshold. After extensive experimental verification, the value of T is more suitable between 0.1 and 0.3.

[0036] The maximum pooling operation in step seven of this invention is specifically shown in the following formula:

[0037] R = max{R K K = 3, 5, 7, 9 (4)

[0038] Where R is the final threshold segmentation result, such as Figure 4 As shown, a to d represent the thresholding results of the traditional method, e to h represent the segmentation results of the method in this paper, and R... K This is a threshold segmentation image at scale K.

[0039] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Those skilled in the art can make various modifications and adjustments within the technical scope disclosed in the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A threshold segmentation method for images after infrared small target background suppression, characterized in that, Includes the following steps: Step 1: Generate a size of Gaussian filter template and utilize Histogram distribution generates one-dimensional vectors ; Step 2: Normalize the input image get , using size sliding window traversal ,get Image blocks Let i = 1, 2, ..., m, respectively according to Histogram generation of image patches one-dimensional vectors , i=1,2,...,m; Step 3: Calculate using cosine similarity respectively and similarity between And binarize the image. get Threshold segmentation map at different scales The binarization operation is shown in the following formula: (3) in, In scale Next The binarization result of each pixel. The set threshold; Step 4: Calculate the threshold segmentation map at four scales. The final threshold segmentation result is obtained by using max pooling. .

2. The threshold segmentation method for an image after background suppression of small infrared targets according to claim 1, characterized in that, The use of Histogram distribution generates one-dimensional vectors For: based on the Gaussian filter template Obtain its histogram and the histogram Divided into There are 10 regions, each containing 10 consecutive gray levels. ; for the histogram of Summing the gray values ​​within each region yields... This data will Each data point forms a one-dimensional vector. .

3. The threshold segmentation method for an image after background suppression of small infrared targets according to claim 1, characterized in that, Step two specifically involves: Step 2.1, Normalize the input image get ; Step 2.2, using a size of The sliding window traverses the image pixel by pixel. ,get Infrared image block , i=1,2,...,m; Step 2.3: Obtain each input infrared image block separately. histogram Let i = 1, 2, ..., m, and then plot the histogram. Divided into There are several regions, each of which is continuous. One gray level, Step 2.4, histogram for each image patch ,right of Summing the gray values ​​within each region yields... This data will Each data point forms a one-dimensional vector. , i=1,2...,m.

4. The threshold segmentation method for an image after background suppression of small infrared targets according to claim 1, characterized in that, and The value is 4.

5. The threshold segmentation method for an image after background suppression of small infrared targets according to claim 1, characterized in that, Step four specifically involves: [Regarding...] By taking four different values, the above seven steps are repeated to obtain four threshold segmentation results. ,right Perform max pooling to obtain the final threshold segmentation result. .

6. The threshold segmentation method for an image after background suppression of small infrared targets according to claim 1, characterized in that, The specific calculation method for the cosine similarity is shown in the following formula: (1) in, For the first Each pixel corresponds to an image block. With Gaussian filter template similarity, For the first Histogram vectors of image patches, Let be the histogram vector of the Gaussian filter template. This represents the total number of pixels in the image.

7. A threshold segmentation method for an image after background suppression of small infrared targets according to claim 1, characterized in that, The normalization operation is specifically shown in the following formula: (2) in, The image after normalization. For the input image, and These are the minimum and maximum pixel values ​​of the input image, respectively.

8. A threshold segmentation method for an image after background suppression of small infrared targets according to claim 1, characterized in that, The value of is in between.

9. A threshold segmentation method for an image after background suppression of small infrared targets according to claim 1, characterized in that, The max pooling operation is shown in the following formula: (4)。

Images