An infrared small target detection method based on improved region growing algorithm

Abstract

The present application relates to small target detection method, specifically to a kind of infrared small target detection method based on improved region growing algorithm, solve the problem that existing region growing algorithm when being applied in infrared small target detection field, long time exists and will confuse target and background area.This application first carries out pre-processing to original infrared image by calculating local contrast, then the obtained pre-processing saliency image is thresholded to screen candidate seed point, effectively avoid the region growing to each pixel in image and waste a lot of time, improve detection efficiency;In order to further enhance infrared small target and suppress background, an improved region growing algorithm is designed, the difference between the infrared small target to be detected and other candidate seed points is enlarged, the confusion of target and background area is avoided, and two features of region coverage ratio and adaptive image entropy difference are designed, to further improve detection accuracy and reduce false alarm rate.

Description

Technical Field

[0001] This invention relates to methods for detecting small targets, specifically to an infrared method for detecting small targets based on an improved region growing algorithm. Background Technology

[0002] Infrared small target detection is an important research area in the field of infrared detection. Due to its advantages such as all-weather operation, strong concealment, and ability to help determine the nature and state of targets, it is widely used in detection systems. However, in practical applications, the distance between the detection system and the target is relatively large, and the images of the target are often small in size, have variable shapes, low grayscale values, and poor contrast due to atmospheric scattering, refraction, and optical defocusing. This presents certain difficulties for the detection task.

[0003] Early infrared small target detection algorithms mostly achieved target enhancement and background suppression through linear or nonlinear filtering. However, these algorithms are prone to generating a large number of false alarms when facing complex backgrounds, which is detrimental to the detection of real infrared small targets. Besides filtering, many researchers have implemented binary classification of background and infrared small targets by introducing Markov random field models. However, when prior information is insufficient, the performance of these algorithms is difficult to guarantee. Other researchers have applied matrix factorization and tensor factorization theories to infrared small target detection. This type of algorithm typically requires multiple iterations of matrix factorization, thus its real-time performance needs further improvement.

[0004] In recent years, target detection algorithms based on human visual mechanisms have attracted widespread attention from researchers due to their good detection performance and real-time performance. In infrared images, there is generally a certain contrast information between the target and the surrounding background within the local neighborhood. Therefore, compared to brightness information, local contrast information is more suitable for infrared small target detection. This type of algorithm typically exhibits good detection performance and real-time performance in scenes where the background changes slowly. However, when there are high-brightness non-target interferences in the scene, such as bright edges and complex high-brightness backgrounds, it cannot effectively suppress the interference. Furthermore, for targets of different sizes, this type of algorithm requires traversing sub-windows of different sizes, and there is room for further improvement in real-time performance.

[0005] Currently, many scholars have begun to apply region growing algorithms to the field of infrared small target detection. However, traditional region growing algorithms directly select candidate seed points in the original infrared image, which consumes a lot of time when growing regions for each pixel in the image. Furthermore, traditional region growing algorithms use each grown point as a seed point for growth discrimination, and use the seed point after each growth as the standard for determining whether to perform growth next time. This can lead to confusion between target and background regions. Summary of the Invention

[0006] The purpose of this invention is to solve the technical problems of existing region growing algorithms being time-consuming and confusing the target with the background when applied to the field of infrared small target detection, and to provide an infrared small target detection method based on an improved region growing algorithm.

[0007] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0008] An infrared small target detection method based on an improved region growing algorithm is characterized by the following steps:

[0009] S1. Acquire raw infrared images of complex scenes;

[0010] S2. Perform image preprocessing on the original infrared image to obtain a preprocessed saliency image;

[0011] S3. Threshold the preprocessed saliency image to select n candidate seed points;

[0012] S4. Use an improved region growing algorithm to grow each candidate seed point on the original infrared image to obtain the growth trajectory of each candidate seed point.

[0013] S5. Calculate the region coverage ratio and adaptive image entropy difference for each candidate seed point based on the growth trajectory of the candidate seed points;

[0014] S6. The confidence level of the candidate seed point is calculated by the region coverage ratio and the adaptive image entropy difference. Then, the candidate seed point and the background are separated by the adaptive threshold method. Finally, the infrared small target to be detected is obtained, and the infrared small target detection based on the improved region growing algorithm is completed.

[0015] Furthermore, step S2 specifically includes:

[0016] S2.1 Use a double-layer sliding window, where the center area of ​​the central window represents the target area, and the other eight neighboring windows represent the background area;

[0017] S2.2 Calculate the mean gray value m of the background of the eight neighboring windows. i

[0018]

[0019] Where: N i This represents the number of pixels in the background of the i-th neighboring window. Let represent the grayscale value of the j-th pixel in the background of the i-th neighboring window;

[0020] S2.3 Calculate the contrast between the central window and neighboring windows.

[0021]

[0022] Where: L n This indicates the maximum grayscale value of the center window;

[0023] S2.4. Preprocess the original infrared image using a local contrast algorithm.

[0024]

[0025] Where: C n This represents the local contrast of pixels in the original infrared image;

[0026] After calculating the local contrast of each pixel in the original infrared image, the preprocessed saliency image PI is obtained.

[0027] Furthermore, step S3 specifically includes:

[0028] S3.1. Threshold the preprocessed saliency image PI and select candidate seed points based on contrast features.

[0029]

[0030] Among them: seeds i (x,y) represents the coordinates of the i-th candidate seed point, and P(x,y) represents the local contrast value of the point (x,y) in the preprocessed saliency image;

[0031] S3.2. Set a protection area of ​​13×13 pixels centered on each candidate seed point. Pixels within the protection area do not participate in the subsequent candidate seed point selection. Return to step S3.1 and continue until n candidate seed points are selected.

[0032] Furthermore, step S4 specifically includes:

[0033] S4.1 Map the location information of the candidate seed points to the corresponding coordinates in the original infrared image to perform region growing;

[0034] S4.2 Calculate candidate regions that meet the growth conditions

[0035]

[0036] Where, k th For threshold parameters, The growth trajectory of the i-th candidate seed point after the l-th growth. express The neighborhood region of the middle pixel, I(s) i) represents the gray value of the i-th candidate seed point, and I(x,y) represents the gray value of the pixel with coordinates (x,y).

[0037] S4.3 From candidate regions that meet the growth conditions Select the pixel with the highest grayscale value (x) * ,y * As a growth point, it is incorporated into the growth trajectory. middle

[0038]

[0039] S4.4 Update the growth trajectory of the i-th candidate seed point after the (l+1)-th growth.

[0040]

[0041] S4.5, when the growth trajectory The number of pixels in the candidate region is greater than or equal to 0.12% of the total number of pixels in the entire original infrared image. When a pixel in the seed does not meet the growth conditions, growth stops, and the growth trajectory of each candidate seed point is finally obtained.

[0042] Furthermore, step S5 specifically includes:

[0043] S5.1 Calculate the center coordinates of the growth trajectory of each candidate seed point.

[0044]

[0045] Among them, A i This represents the growth trajectory after the i-th candidate seed point has completed its growth.

[0046] S5.2, with Centered on a i Generate a rectangular region Z with side length Z. i

[0047]

[0048] Where: C i This represents the number of pixels in the growth trajectory of the i-th candidate seed point;

[0049] S5.3 Calculate Z i With A i Intersecting pixels h i (x)

[0050] h i (x)={x|x∈Z i ∩Ai};

[0051] S5.4 Calculate the region coverage ratio (RCR) of the i-th candidate seed point. i

[0052]

[0053] Where, m i h i The number of pixels in (x);

[0054] S5.5, divide the rectangular region Z i Mapped onto the preprocessed saliency image PI as an inner window W1;

[0055] S5.6, Set the size to (a) i +v)×(a i The outer window W2 of the +v) is used as the background area;

[0056] Where v represents a constant;

[0057] The transition window W3 between the outer window W2 and the inner window W1 is calculated.

[0058] W3 = W2 - W1;

[0059] S5.7 Calculate the adaptive image entropy difference (IED) of the i-th candidate seed point. i

[0060] IED i =ImgEntrogy(W3)-ImgEntrogy(W2)

[0061] Where: ImgEntrogy(W3) represents the image entropy of the transition window W3, and ImgEntrogy(W2) represents the image entropy of the outer window W2.

[0062] Furthermore, step S6 specifically includes:

[0063] S6.1, through the area coverage ratio (RCR) i Adaptive Image Entropy Difference (IED) i Calculate the confidence score CON for each candidate seed point. i

[0064] CON i =RCR i ×IED i ;

[0065] S6.2 Calculate the confidence level CON of the candidate seed point. i The segmentation threshold th separates the candidate seed points from the background.

[0066] th=μ+kσ

[0067] Where μ and σ represent the confidence scores CON of the n candidate seed points, respectively. i The mean and standard deviation, where k is a constant;

[0068] Finally, the infrared small target to be detected is obtained, and the infrared small target detection based on the improved region growing algorithm is completed.

[0069] Compared with the prior art, the beneficial effects of the present invention are:

[0070] This invention provides an infrared small target detection method based on an improved region growing algorithm. First, the original infrared image is preprocessed by calculating local contrast. Then, the preprocessed saliency image is thresholded to filter candidate seed points, effectively avoiding the waste of time by performing region growing on every pixel in the image and improving detection efficiency. To further enhance the infrared small target and suppress the background, an improved region growing algorithm is designed, which expands the difference between the infrared small target to be detected and other candidate seed points, avoiding confusion between the target and the background region. Two features, region coverage ratio and adaptive image entropy difference, are designed to further improve detection accuracy and reduce false alarm rate. Attached Figure Description

[0071] Figure 1 This is a flowchart of an embodiment of an infrared small target detection method based on an improved region growing algorithm according to the present invention;

[0072] Figure 2 In this embodiment of the invention, the original infrared image and its three-dimensional image obtained in step S1 are shown; where a is the original infrared image and b is the three-dimensional image of the original infrared image.

[0073] Figure 3 In this embodiment of the invention, the preprocessed saliency image and its three-dimensional image obtained in step S2 are shown; where a is the preprocessed saliency image and b is the three-dimensional image of the preprocessed saliency image.

[0074] Figure 4 In this embodiment of the invention, the candidate seed point image selected in step S3 is shown.

[0075] Figure 5 This is the growth trajectory image of the candidate seed points obtained in step S4 in this embodiment of the invention;

[0076] Figure 6 This is a schematic diagram of the double-layer region sliding window in step S2.1 of this embodiment of the invention;

[0077] Figure 7This is a schematic diagram of the inner window, outer window, and transition window in an embodiment of the present invention;

[0078] Figure 8 This is a comparison image of the original infrared image and the infrared small target to be detected in an embodiment of the present invention; where a is the original infrared image and b is the image of the infrared small target to be detected. Detailed Implementation

[0079] The present invention will be further described below with reference to the accompanying drawings and exemplary embodiments.

[0080] Reference Figures 1-8 The present invention provides an infrared small target detection method based on an improved region growing algorithm, comprising the following steps:

[0081] S1. Acquire raw infrared images of complex scenes;

[0082] like Figure 2 As shown in figure a, the original infrared image was obtained. Figure 2 b is its corresponding three-dimensional image.

[0083] S2. Perform image preprocessing on the original infrared image to obtain a preprocessed saliency image;

[0084] like Figure 3 As shown in figure a, this is the preprocessed saliency image. Figure 3 b is its corresponding three-dimensional image;

[0085] Step S2 is as follows:

[0086] S2.1 Use a double-layer sliding window, where the center area of ​​the central window represents the target area, and the other eight neighboring windows represent the background area;

[0087] A schematic diagram of a dual-layer sliding window is shown below. Figure 6 As shown, T0 represents the center window, and B1-B8 represent the neighboring windows;

[0088] S2.2 Calculate the mean gray value m of the background of the eight neighboring windows. i

[0089]

[0090] Where: N i This represents the number of pixels in the background of the i-th neighboring window. Let represent the grayscale value of the j-th pixel in the background of the i-th neighboring window;

[0091] S2.3 Calculate the contrast between the central window and neighboring windows.

[0092]

[0093] Where: L n This indicates the maximum grayscale value of the center window;

[0094] S2.4. Preprocess the original infrared image using a local contrast algorithm.

[0095]

[0096] Where: C n This represents the local contrast of pixels in the original infrared image;

[0097] After calculating the local contrast of each pixel in the original infrared image, the preprocessed saliency image PI is obtained.

[0098] S3. Threshold the preprocessed saliency image to select n candidate seed points;

[0099] Figure 4 Images of the selected candidate seed points;

[0100] Step S3 is as follows:

[0101] S3.1. Threshold the preprocessed saliency image PI and select candidate seed points based on contrast features.

[0102]

[0103] Among them: seeds i (x,y) represents the coordinates of the i-th candidate seed point, and P(x,y) represents the local contrast value of the point (x,y) in the preprocessed saliency image;

[0104] S3.2. Set a protection area of ​​13×13 pixels centered on each candidate seed point. Pixels within the protection area do not participate in the subsequent candidate seed point selection. Return to step S3.1 and continue until n candidate seed points are selected.

[0105] S4. Use an improved region growing algorithm to grow each candidate seed point on the original infrared image to obtain the growth trajectory of each candidate seed point.

[0106] Figure 5 The growth trajectory image of the candidate seed points;

[0107] Step S4 is as follows:

[0108] S4.1 Map the location information of the candidate seed points to the corresponding coordinates in the original infrared image to perform region growing;

[0109] S4.2 Calculate candidate regions that meet the growth conditions

[0110]

[0111] Where, k th For threshold parameters, The growth trajectory of the i-th candidate seed point after the l-th growth. express The neighborhood region of the middle pixel, I(s) i ) represents the gray value of the i-th candidate seed point, and I(x,y) represents the gray value of the pixel with coordinates (x,y).

[0112] S4.3 From candidate regions that meet the growth conditions Select the pixel with the highest grayscale value (x) * ,y * As a growth point, it is incorporated into the growth trajectory. middle

[0113]

[0114] S4.4 Update the growth trajectory of the i-th candidate seed point after the (l+1)-th growth.

[0115]

[0116] S4.5, when the growth trajectory The number of pixels in the candidate region is greater than or equal to 0.12% of the total number of pixels in the entire original infrared image. When a pixel in the seed does not meet the growth conditions, growth stops, and the growth trajectory of each candidate seed point is finally obtained.

[0117] S5. Calculate the region coverage ratio and adaptive image entropy difference for each candidate seed point based on the growth trajectory of the candidate seed points;

[0118] Step S5 is as follows:

[0119] S5.1 Calculate the center coordinates of the growth trajectory of each candidate seed point.

[0120]

[0121] Among them, A i This represents the growth trajectory after the i-th candidate seed point has completed its growth.

[0122] S5.2, with Centered on a i Generate a rectangular region Z with side length Z.i

[0123]

[0124] Where: C i This represents the number of pixels in the growth trajectory of the i-th candidate seed point;

[0125] S5.3 Calculate Z i With A i Intersecting pixels h i (x)

[0126] h i (x)={x|x∈Z i ∩A i};

[0127] S5.4 Calculate the region coverage ratio (RCR) of the i-th candidate seed point. i

[0128]

[0129] Where, m i h i The number of pixels in (x);

[0130] S5.5, divide the rectangular region Z i Mapped onto the preprocessed saliency image PI as an inner window W1;

[0131] S5.6, Set the size to (a) i +v)×(a i The outer window W2 of the +v) is used as the background area;

[0132] Where v represents a constant;

[0133] The transition window W3 between the outer window W2 and the inner window W1 is calculated.

[0134] W3 = W2 - W1;

[0135] A schematic diagram of the outer window W2, the inner window W1, and the transition window W3 is shown below. Figure 7 As shown, the outermost box is the outer window W2, the center box is the inner window W1, and the blue area between them is the transition window W3;

[0136] S5.7 Calculate the adaptive image entropy difference (IED) of the i-th candidate seed point. i

[0137] IED i =ImgEntrogy(W3)-ImgEntrogy(W2)

[0138] Where: ImgEntrogy(W3) represents the image entropy of the transition window W3, and ImgEntrogy(W2) represents the image entropy of the outer window W2.

[0139] S6. The confidence of the candidate seed points is calculated by the region coverage ratio and the adaptive image entropy difference. Then, the candidate seed points and the background are separated by the adaptive threshold method. Finally, the infrared small target to be detected is obtained, and the infrared small target detection based on the improved region growing algorithm is completed.

[0140] Step S6 is as follows:

[0141] S6.1, through the area coverage ratio (RCR) i Adaptive Image Entropy Difference (IED) i Calculate the confidence score CON for each candidate seed point. i

[0142] CON i =RCR i ×IED i ;

[0143] S6.2 Calculate the confidence level CON of the candidate seed point. i The segmentation threshold th separates the candidate seed points from the background.

[0144] th=μ+kσ

[0145] Where μ and σ represent the confidence scores CON of the n candidate seed points, respectively. i The mean and standard deviation of the , where k is a constant, and in this embodiment, k is 2;

[0146] Finally, the infrared small target to be detected is obtained, and the infrared small target detection based on the improved region growing algorithm is completed.

[0147] A comparison image of the original infrared image and the infrared small target to be detected after detection by the infrared small target detection method based on the improved region growing algorithm provided by this invention is shown below. Figure 8 As shown, where, Figure 8 a is the original infrared image. Figure 8 Image b shows the infrared small target to be detected. As can be seen, this method can accurately detect the infrared small target and precisely segment it from the background.

[0148] The embodiments described above are merely illustrative of specific implementations of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for detecting small infrared targets based on an improved region growing algorithm, characterized in that, Includes the following steps: S1. Acquire raw infrared images of complex scenes; S2. Perform image preprocessing on the original infrared image to obtain a preprocessed saliency image; S3. Threshold the preprocessed saliency image to select n candidate seed points; S4. Use an improved region growing algorithm to grow each candidate seed point on the original infrared image to obtain the growth trajectory of each candidate seed point. S5. Calculate the region coverage ratio and adaptive image entropy difference for each candidate seed point based on the growth trajectory of the candidate seed points; S6. The confidence level of the candidate seed point is calculated by the region coverage ratio and the adaptive image entropy difference. Then, the candidate seed point and the background are separated by the adaptive threshold method. Finally, the infrared small target to be detected is obtained, and the infrared small target detection based on the improved region growing algorithm is completed.

2. The infrared small target detection method based on the improved region growing algorithm according to claim 1, characterized in that, Step S2 is as follows: S2.1 Use a double-layer sliding window, where the center area of ​​the central window represents the target area, and the other eight neighboring windows represent the background area; S2.2 Calculate the mean gray value m of the background of the eight neighboring windows. i Where: N i This represents the number of pixels in the background of the i-th neighboring window. Let represent the grayscale value of the j-th pixel in the background of the i-th neighboring window; S2.3 Calculate the contrast between the central window and neighboring windows. Where: L n This indicates the maximum grayscale value of the center window; S2.

4. Preprocess the original infrared image using a local contrast algorithm. Where: C n This represents the local contrast of pixels in the original infrared image; After calculating the local contrast of each pixel in the original infrared image, the preprocessed saliency image PI is obtained.

3. The infrared small target detection method based on the improved region growing algorithm according to claim 2, characterized in that, Step S3 is as follows: S3.

1. Threshold the preprocessed saliency image PI and select candidate seed points based on contrast features. Among them: seeds i (x,y) represents the coordinates of the i-th candidate seed point, and P(x,y) represents the local contrast value of the point (x,y) in the preprocessed saliency image; S3.

2. Set a protection area of ​​13×13 pixels centered on each candidate seed point. Pixels within the protection area do not participate in the subsequent candidate seed point selection. Return to step S3.1 and continue until n candidate seed points are selected.

4. The infrared small target detection method based on the improved region growing algorithm according to claim 3, characterized in that, Step S4 is as follows: S4.1 Map the location information of the candidate seed points to the corresponding coordinates in the original infrared image to perform region growing; S4.2 Calculate candidate regions that meet the growth conditions Where, k th For threshold parameters, The growth trajectory of the i-th candidate seed point after the l-th growth. express The neighborhood region of the middle pixel, I(s) i ) represents the gray value of the i-th candidate seed point, and I(x,y) represents the gray value of the pixel with coordinates (x,y). S4.3 From candidate regions that meet the growth conditions Select the pixel with the highest grayscale value (x) * ,y * As a growth point, it is incorporated into the growth trajectory. middle S4.4 Update the growth trajectory of the i-th candidate seed point after the (l+1)-th growth. S4.5, when the growth trajectory The number of pixels in the candidate region is greater than or equal to 0.12% of the total number of pixels in the entire original infrared image. When a pixel in the seed does not meet the growth conditions, growth stops, and the growth trajectory of each candidate seed point is finally obtained.

5. The infrared small target detection method based on the improved region growing algorithm according to claim 4, characterized in that, Step S5 is as follows: S5.1 Calculate the center coordinates of the growth trajectory of each candidate seed point. Among them, A i This represents the growth trajectory after the i-th candidate seed point has completed its growth; S5.2, with Centered on a i Generate a rectangular region Z with side length Z. i Where: C i This represents the number of pixels in the growth trajectory of the i-th candidate seed point; S5.3 Calculate Z i With A i Intersecting pixels h i (x) S5.4 Calculate the region coverage ratio (RCR) of the i-th candidate seed point. i Where, m i h i The number of pixels in (x); S5.5, divide the rectangular region Z i Mapped onto the preprocessed saliency image PI as an inner window W1; S5.6, Set the size to (a) i +v)×(a i The outer window W2 of the +v) is used as the background area; Where v represents a constant; The transition window W3 between the outer window W2 and the inner window W1 is calculated. W3 = W2 - W1; S5.7 Calculate the adaptive image entropy difference (IED) of the i-th candidate seed point. i IED i =ImgEntrogy(W3)-ImgEntrogy(W2) Where: ImgEntrogy(W3) represents the image entropy of the transition window W3, and ImgEntrogy(W2) represents the image entropy of the outer window W2.

6. The infrared small target detection method based on the improved region growing algorithm according to claim 5, characterized in that, Step S6 is as follows: S6.1, via regional coverage ratio (RCR) i Adaptive Image Entropy Difference (IED) i Calculate the confidence score CON for each candidate seed point. i CON i =RCR i ×IED i ; S6.2 Calculate the confidence level CON of the candidate seed point. i The segmentation threshold th separates the candidate seed points from the background. th=μ+kσ Where: μ and σ represent the confidence scores CON of the n candidate seed points, respectively. i The mean and standard deviation, where k is a constant; Finally, the infrared small target to be detected is obtained, and the infrared small target detection based on the improved region growing algorithm is completed.

Images