Infrared image hierarchical feature enhancement algorithm based on guided image filtering

Abstract

The invention provides an infrared image hierarchical feature enhancement algorithm based on guided image filtering. The method comprises the following steps: S1, decomposing an original infrared image into a base layer and a detail layer by using wavelet decomposition; s2, effective detail layer feature components are obtained by guiding image filtering; s3, fusing the detail layer feature component with the detail layer image information to obtain a detail layer fusion component; s4, performing image enhancement on the detail layer fusion component through an enhancement module to obtain a detail layer image component; and S5, carrying out weighted fusion on the basic layer information and the detail layer image component to obtain a high-contrast infrared image.The infrared image hierarchical enhancement method based on the guided filtering algorithm is provided for solving the technical problems that infrared imaging is weak, the target contrast ratio is low, the edge is fuzzy, the contrast ratio is low, the signal intensity is weak and the like and is prone to being submerged by noise. The infrared image quality can be effectively improved, and the method belongs to the technical field of infrared image processing.

Description

Technical Field

[0001] The present invention belongs to the technical field of infrared image processing, and in particular relates to an infrared image hierarchical feature enhancement algorithm based on guided image filtering. Background Art

[0002] Infrared image enhancement involves a series of technologies aimed at improving the visual effects and analytical performance of infrared images. The goal of infrared image enhancement is to improve the contrast, clarity and target recognition capabilities of the image, especially when the distinction between the target and the background is not high or the image is interfered with by noise. Traditional algorithms include grayscale transformation, histogram equalization, image smoothing filtering and image sharpening filtering. Infrared images are often affected by noise, especially in low temperature and low light environments, where the noise is more obvious and affects the image quality. Infrared images generally have low contrast, and the difference between the target and the background is not obvious, making it difficult to effectively identify. Detailed information is often compressed or blurred, especially when long exposures or long distances are used. Moreover, since infrared images usually have a large dynamic range, improper processing will result in loss of details in overexposed or dark areas.

[0003] Traditional image filtering algorithms, such as mean filtering, Gaussian filtering, and median filtering, while simple and effective, often suffer from common drawbacks: they often lead to blurred edges and loss of detail, especially when complex structures or subtle changes are present in the image. These methods typically employ global smoothing strategies and are unable to adaptively process details across different regions. This means that while they may denoise and enhance the image, they may also destroy important image features. Summary of the Invention

[0004] The purpose of the present invention is to provide an infrared image hierarchical feature enhancement algorithm based on guided image filtering. In order to solve the technical problems that weak targets in infrared imaging have low contrast, blurred edges, low contrast, weak signal strength and are easily submerged by noise, a layered infrared image enhancement method based on guided filtering algorithm is proposed to more effectively improve the quality of infrared images.

[0005] To achieve the above object, the present invention provides an infrared image hierarchical feature enhancement algorithm based on guided image filtering, the method comprising the following steps:

[0006] S1, decompose the original infrared image into a base layer and a detail layer using wavelet decomposition;

[0007] S2, obtain effective detail layer feature components by guiding image filtering;

[0008] S3, fusing the detail layer feature component with the detail layer image information to obtain the detail layer fusion component;

[0009] S4, the detail layer fusion component is subjected to image enhancement through an enhancement module to obtain a detail layer image component;

[0010] S5, the base layer information is weightedly fused with the detail layer image component to obtain a high-contrast infrared image.

[0011] Further, the wavelet decomposition method in the step S1 adopts a Symlet wavelet function to perform one-stage wavelet decomposition, and an image input by a network has a size of 640*512; the collected infrared image is decomposed into a low-frequency part containing more background features and a high-frequency part containing more detail textures and noise information through wavelet transformation.

[0012] Further, the guided image filtering algorithm in the step S2 dynamically adjusts the strength of filtering through a regularization parameter epsilon, and the coefficient a k is a formula

[0013]

[0014] ; wherein p is an input image, refers to a variance;

[0015] Dynamic balance between denoising and edge preservation is realized through dynamic adjustment of the parameter epsilon and the coefficient a k .

[0016] Further, the step S3 fuses the detail layer feature component and the detail layer image information through a formula R(x,y)=a1N(x,y)+a2E(x,y), wherein a1 and a2 are fusion coefficients of N(x,y) and E(x,y) respectively.

[0017] Further, the enhancement module in the step S4 contains a combination operation of open operation, bilateral filtering, histogram equalization and gamma transformation.

[0018] Further, the step S5 realizes weighted fusion through a weighted fusion formula

[0019] I(x,y)=exp{ln[R'(x,y)]+ln[a3Q(x,y)]}, a3 is an intensity coefficient of Q(x,y), and the value range is between 0 and 1.

[0020] The present application provides the infrared image hierarchical feature enhancement algorithm based on guided image filtering, through using wavelet decomposition images, respectively extracting feature information and fusing, then directly weighting the enhanced detail layer with the base layer, so that the enhanced image has rich detail information, effectively enhances the target and suppresses the interference of complex background and noise.

[0021] The present application will be described in detail below in combination with the drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flow chart of the infrared image algorithm based on guided image filtering.

[0023] Figure 2 This is a framework diagram of the infrared image algorithm based on guided image filtering.

[0024] Figure 3 It is the module diagram of the enhanced denoising algorithm.

[0025] Figure 4 It is a graph of algorithm input and output. DETAILED DESCRIPTION

[0026] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose, the specific implementation methods, structural features and effects of the present invention are described in detail below with reference to the accompanying drawings and examples.

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0028] In the description of the present invention, it should be understood that the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "aligned", "overlap", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention.

[0029] The terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, features specified as "first" or "second" may explicitly or implicitly include one or more of such features; and in the description of the present invention, unless otherwise specified, "plurality" means two or more.

[0030] Example 1

[0031] This embodiment provides Figures 1 to 4 The infrared image hierarchical feature enhancement algorithm based on guided image filtering is shown, and the method includes the following steps:

[0032] S1. Use wavelet decomposition to decompose the original infrared image into a base layer and a detail layer. First, use wavelet decomposition to decompose the image to be decomposed and extract the base layer and detail layer from it, obtaining the image's detail layer information component N(x, y) and base layer information component Q(x, y). The base layer contains a large amount of low-frequency information reflecting the image background information, and the detail layer contains a large amount of high-frequency information reflecting the image edge details.

[0033] S2. Obtain effective detail layer feature components through guided image filtering, that is, perform feature enhancement extraction on the detail layer through guided filtering to obtain detail layer feature components E(x, y) to facilitate subsequent image fusion and processing;

[0034] S3. Fusing the detail layer feature component with the detail layer image information to obtain a detail layer fusion component; that is, fusing the detail layer information component N(x,y) with the detail layer feature component E(x,y) in an appropriate ratio to obtain a detail layer fusion component R(x,y). By fusing the detail layer information component N(x,y) with the detail layer feature component E(x,y), the necessary details of the image can be retained while highlighting the enhanced target features.

[0035] S4. The detail layer fusion component is enhanced by the enhancement module to obtain the detail layer image component; the detail layer fusion component R(x, y) obtained after fusion is subjected to image enhancement processing to obtain the detail layer image component R′(x, y). The image is enhanced by the enhancement module, and morphological image enhancement, filtering, sharpening and other operations are performed on the detail layer to enhance the target edge information, achieve image smoothing and denoising performance, enhance the contrast between the target pixels and the background pixels in the image, improve the dynamic difference of the grayscale values ​​between the pixels in the image, make the linear response of the image exposure intensity closer to the response perceived by the human eye, and thus improve the contrast of the image;

[0036] S5. Perform weighted fusion of the base layer information and the detail layer image components to obtain a high-contrast infrared image. Finally, perform weighted fusion of the base layer information component Q(x,y) and the detail layer image component R′(x,y) to obtain a high-contrast infrared image, thereby enhancing the original infrared image. The fusion ratio is adjusted by the weight coefficient so that the feature information is prominent and does not overflow, thereby obtaining a high-contrast infrared image I(x,y), thereby ultimately enhancing the target original image.

[0037] Furthermore, in step S1, the wavelet decomposition method uses the Symlet wavelet function to perform first-level wavelet decomposition. The network input image size is 640×512, and the collected infrared image is decomposed into a low-frequency portion containing more background features and a high-frequency portion containing more detailed texture and noise information through wavelet transform. The Symlet wavelet reduces edge distortion by virtue of its approximate symmetry, and combines high regularity and high vanishing moment to suppress noise while improving the smoothness of the base layer. It is an efficient choice for balancing detail preservation and noise suppression in hierarchical feature enhancement of infrared images.

[0038] Wavelet decomposition methods are commonly used in image processing. Wavelet transform can decompose an image into different resolution levels, thereby better capturing detailed information in the image. At low-resolution levels, it mainly manifests as larger structures and changes, while at high-resolution levels, it can capture subtle textures and noise in the image. Combining this multi-scale information with guided filtering can more accurately enhance key features in the image, suppressing noise while enhancing image contrast.

[0039] Furthermore, in step S2, the image filtering algorithm is guided to dynamically adjust the filtering strength through the regularization parameter ε, and the coefficient a k The formula is

[0040] By dynamically adjusting the parameter ε and the coefficient a k Achieve dynamic balance between denoising and edge preservation;

[0041] The formula based on guided image filtering (GIF) is derived as follows:

[0042] p i =q i +e i

[0043] (1),

[0044] Among them, P i is the original image, q i is the image contour information, e i Image details and noise information;

[0045] The base layer q i It is defined as a model that is k There is a guide image I i Local linear model of q i and the guide image I i There is a local linear relationship between:

[0046]

[0047] a k and bk is the coefficient, w k is the window; for a deterministic window w with a radius r k ,(a k ,b k ) will also be the only deterministic constant coefficients, and their values ​​can be obtained by minimizing the cost function;

[0048] The cost function of this algorithm is as follows:

[0049]

[0050] Among them, ε is the minimum regularization parameter of the loss function, which can prevent the coefficient a k Too large, take the derivative and you get:

[0051]

[0052] b k =p k -a k μ k (5),

[0053] where p k and μ k are the average values ​​within P and I windows respectively, for a k The solution can be obtained by formula (6):

[0054]

[0055] Finally, we get the coefficient a k :

[0056]

[0057] Where p is the input image, Refers to the variance.

[0058] When ε=0, a=1, b=0 is E(a k ,b k ), the guided filter effect is not obvious, and the output is equal to the input; when ε>0, a≈0, b=pk corresponds to a weighted mean filter; when ε changes greatly, a→1 and b→0 can better maintain the image edge, which has the same effect as the weighted mean filter; when ε changes in the defined area, the window ω k The size remains unchanged, and it has a better effect on image edge and detail processing.

[0059] Furthermore, step S3 fuses the detail layer feature components with the detail layer image information using the following formula: R(x,y)=α1N(x,y)+α2E(x,y), where α1 and α2 are the fusion coefficients of N(x,y) and E(x,y), respectively. By adjusting the fusion ratio of the base layer and the detail layer, i.e., the fusion coefficients α1 and α2, information overflow or contrast imbalance can be avoided.

[0060] In conventional operations, using the detail layer information component N(x,y) alone will introduce high-frequency noise, and using the detail layer feature component E(x,y) alone may lose some of the original details of the image. By adjusting the fusion coefficient to generate the detail layer fusion component R(x,y), the enhanced target detail feature display is highlighted while retaining the necessary details of the image.

[0061] In the fusion formula R(x,y)=α1N(x,y)+α2E(x,y) (10):

[0062] The fusion coefficients of N(x,y) and E(x,y) are α1 and α2 respectively, which determine the proportion of the two components when they are fused. E(x,y) represents the pixel value of the feature component extracted through the detail layer. Each pixel contains a large amount of detail feature information. In order to ensure that the most detail feature information is obtained after extraction, the value of α1 is in the range of 0-1. N(x,y) represents the pixel value of the target detail feature component. The maximum grayscale information obtained is 50. The value range of α2 is 3-5 to ensure that the information does not overflow.

[0063] Further, such as Figure 3 As shown, the enhancement module in step S4 includes a combination of opening operation, bilateral filtering, histogram equalization and gamma transformation;

[0064] The enhancement of detail layer is mainly achieved through morphological image enhancement algorithm, which mainly uses the combination of dilation and erosion methods to achieve effective extraction and denoising of target feature information;

[0065] Dilation and erosion are two basic operations in morphological image processing, which are widely used in image analysis, edge detection, denoising, image restoration and other tasks. These two operations change the shape of the image by performing local operations on the pixel structure of the image. They are mainly based on morphological structural elements (usually a small binary template) to operate on the target area in the image as Z 2 Sets A and B in the erosion are represented as The corrosion of B on A is defined as:

[0066]

[0067] As Z2 The sets A and B in the expansion are expressed as The expansion of B to A is defined as:

[0068]

[0069] Opening is an important basic operation in morphological image processing. It is composed of erosion and dilation. It is mainly used to remove small objects and noise in the image while preserving the overall shape of the object.

[0070] For infrared images containing a large amount of noise information, the bilateral filtering algorithm is used. Bilateral filtering is a nonlinear filtering method that aims to simultaneously consider the spatial proximity and pixel value similarity of the image, thereby achieving denoising while retaining edge information;

[0071] The histogram equalization enhancement algorithm is used to enhance the contrast between target pixels and background pixels in an image and improve the dynamic difference in grayscale values ​​between pixels in the image. Histogram equalization stretches the grayscale distribution of the image to make the grayscale histogram of the image close to a uniform distribution, thereby improving the brightness contrast of low-contrast images and making details clearer.

[0072] Finally, the enhanced image is sharpened by gamma transform to improve the image contrast. Gamma transform adjusts the overall brightness and contrast of the image by performing a power transformation on the pixel values ​​of the image, and finally obtains the detail layer image component R′(x,y).

[0073] Furthermore, the weighted fusion in step S5 is implemented by the following weighted fusion formula:

[0074] I(x,y)=exp{ln[R′(x,y)]+ln[α3Q(x,y)]}, where α3 is the intensity coefficient of Q(x,y), ranging from 0 to 1;

[0075] In formula (11), α3 is used to adjust the intensity of the component images in the final output image I(x, y), ranging from 0 to 1, to improve the signal-to-noise ratio (SNR) and edge clarity of the final image and reduce the background noise of the image.

[0076] The present invention first uses wavelet decomposition to decompose the infrared image into a base layer and a detail layer, extracts and fuses the feature information on different layers, performs morphological image enhancement, filtering, sharpening and other operations on the detail layer, strengthens the target edge information, achieves image smoothing and denoising performance, enhances the contrast between target pixels and background pixels in the image, improves the dynamic difference in grayscale values ​​between pixels in the image, and makes the linear response of the image exposure intensity closer to the response perceived by the human eye, thereby improving the image contrast;

[0077] Then the effective component of the target is extracted using a guided filter algorithm, and then the image information of the detail layer is fused with the extracted effective component and enhanced, and the weight coefficient is adjusted in the fusion process, so that the feature information is prominent and does not overflow, and finally a high-contrast infrared image is obtained;

[0078] Finally, the image information of the base layer is weighted and fused with the enhanced detail information obtained in the foregoing, so that the target is enhanced;

[0079] Through multi-module cooperation and weight optimization, the contradiction between noise amplification and detail loss in the traditional method is further solved, and the target recognition capability of the infrared image is improved.

[0080] The above is a further detailed description of the present application in combination with specific preferred embodiments, and the specific implementation of the present application cannot be limited to these descriptions. For ordinary skilled persons in the technical field to which the present application belongs, a number of simple deductions or substitutions can be made without departing from the concept of the present application, and all of them should be regarded as falling within the protection scope of the present application.

Claims

1. A hierarchical feature enhancement algorithm for infrared images based on guided image filtering, characterized in that: The method comprises the following steps: S1, decompose the original infrared image into a base layer and a detail layer using wavelet decomposition; S2, obtain effective detail layer feature components by guiding image filtering; S3, fusing the detail layer feature component with the detail layer image information to obtain the detail layer fusion component; S4, the detail layer fusion component is enhanced through the enhancement module to obtain the detail layer image component; S5. Perform weighted fusion of the base layer information and the detail layer image components to obtain a high-contrast infrared image.

2. The infrared image hierarchical feature enhancement algorithm based on guided image filtering according to claim 1, characterized in that: In step S1, the wavelet decomposition method uses the Symlet wavelet function to perform first-level wavelet decomposition. The image size of the network input is 640×512, and the collected infrared image is decomposed into a low-frequency part containing more background features and a high-frequency part containing more detailed texture and noise information through wavelet transform.

3. The infrared image hierarchical feature enhancement algorithm based on guided image filtering according to claim 1, characterized in that: In step S2, the image filtering algorithm is guided to dynamically adjust the filtering strength through the regularization parameter ε, and the coefficient a k The formula is ； Where p is the input image, refers to the variance; By dynamically adjusting the parameter ε and the coefficient a k Achieve dynamic balance between denoising and edge preservation.

4. The infrared image hierarchical feature enhancement algorithm based on guided image filtering according to claim 1, characterized in that: The step S3 is implemented by fusing the detail layer feature component with the detail layer image information through the following formula: R(x,y)=α1N(x,y)+α2E(x,y), where α1 and α2 are the fusion coefficients of N(x,y) and E(x,y) respectively.

5. The infrared image hierarchical feature enhancement algorithm based on guided image filtering according to claim 1, characterized in that: The enhancement module in step S4 includes a combination of opening operation, bilateral filtering, histogram equalization and gamma transformation.

6. The infrared image hierarchical feature enhancement algorithm based on guided image filtering according to claim 1, characterized in that: The weighted fusion in step S5 is implemented by the following weighted fusion formula: I(x,y)=exp{ln[R′(x,y)]+ln[α3Q(x,y)]}, where α3 is the intensity coefficient of Q(x,y), and its value range is between 0 and 1.

Images