High dynamic infrared image detail enhancement method, system and computer storage medium

Abstract

The application discloses a high-dynamic infrared image detail enhancement method, system and computer storage medium, and the method comprises the following steps: carrying out denoising processing on a high-dynamic infrared original image by using guided filtering to obtain a denoised image; carrying out image layering processing on the denoised image by using Gaussian filtering to obtain a low-frequency image, and obtaining a high-frequency image by using the low-frequency image and the high-dynamic infrared original image; obtaining a contrast enhancement function according to human eye visual characteristics; carrying out enhancement processing on the high-frequency image by using the contrast enhancement function to obtain an enhanced detail image; carrying out brightness-preserving double-platform histogram dimming processing on the high-dynamic infrared original image to obtain a dimmed image; and carrying out weighted fusion on the dimmed image and the enhanced detail image. The application has strong denoising effect, strong scene adaptability, can enhance the intensity of weak details while protecting original clear edges, and can effectively reduce the pseudo-edge phenomenon.

Description

Technical Field

[0001] This invention relates to the field of infrared image detail enhancement technology, and in particular to a high dynamic range infrared image detail enhancement method, system, and computer storage medium. Background Technology

[0002] Infrared thermal imaging systems, due to their unique temperature difference imaging method, can provide video images even in harsh conditions such as darkness and heavy fog, and are widely used in military reconnaissance, industrial production, and other fields. Currently, infrared sensors can output infrared images with a bit width of up to 14 bits of grayscale, while ordinary display devices can only display infrared images with 8 bits of grayscale. The grayscale range of a 14-bit infrared image far exceeds the response range of ordinary display devices, hence such images are called high dynamic range (HDR) images. To allow real-time observation of infrared images on display devices, the dynamic range of the original 14-bit infrared image must be compressed to 8 bits. For scenes with gradual energy changes, the grayscale distribution of the generated image will be concentrated in a narrow range, while for scenes with drastic energy changes, the grayscale will be scattered over a wider range. For HDR infrared image processing, the key is to convert the information in the 14-bit image into an 8-bit image that is observable to the human eye while preserving the original image's detail and providing good contrast for human observation.

[0003] Existing infrared image enhancement techniques include dynamic range segmentation algorithms based on bilateral filtering. However, bilateral filters are prone to gradient flipping at edges where image grayscale changes are significant, resulting in halo artifacts and false edges. Guided filtering-based DDE algorithms replace bilateral filtering, preserving image details while avoiding gradient flipping. However, these algorithms suffer from poor scene adaptability due to parameter setting. Adaptive infrared image detail enhancement algorithms based on guided filtering use histogram distribution information to determine adaptive thresholds in the base layer image, removing invalid grayscale values ​​and allowing the base layer image to better display valid information. However, these algorithms have poor denoising performance; when the image contains a large area of ​​sky background, the processed image contains significant noise interference. Summary of the Invention

[0004] This invention provides a high dynamic range infrared image detail enhancement method, system, and computer storage medium to solve problems in the prior art such as halo artifacts, false edges, poor noise reduction, and a large amount of noise interference in the processed image when the image contains a large area of ​​sky background.

[0005] On one hand, embodiments of the present invention provide a method for enhancing details in high dynamic range infrared images, including:

[0006] A denoised image is obtained by using guided filtering to denoise the original high dynamic range infrared image.

[0007] The denoised image is processed by Gaussian filtering to obtain a low-frequency image, and a high-frequency image is obtained by using the low-frequency image and the original high-dynamic infrared image.

[0008] The contrast enhancement function is obtained based on the visual characteristics of the human eye;

[0009] The high-frequency image is enhanced using the contrast enhancement function to obtain an enhanced detail image.

[0010] The high dynamic range infrared raw image is subjected to brightness-preserving dual-platform histogram dimming processing to obtain a dimmed image;

[0011] The dimmed image and the enhanced detail image are then weighted and fused.

[0012] In one possible implementation, denoising a high dynamic range infrared (HMR) image using guided filtering to obtain a denoised image includes: determining the weight coefficients and bias coefficients of the guided filtering based on the HMR image and the guided image; and filtering the HMR image based on the guided image, the weight coefficients, and the bias coefficients to obtain the denoised image.

[0013] In one possible implementation, the denoised image is processed by image layering using Gaussian filtering to obtain a low-frequency image, and a high-frequency image is obtained using the low-frequency image and the original high-dynamic infrared image. This includes: setting a Gaussian filter template mask; performing a convolution operation between the Gaussian filter template mask and the original high-dynamic infrared image to obtain the low-frequency image; and subtracting the original high-dynamic infrared image from the low-frequency image to obtain the high-frequency image.

[0014] In one possible implementation, the contrast enhancement function is the product of a local detail adjustment function and a contrast limiting function.

[0015] In one possible implementation, the high-frequency image is enhanced using the contrast enhancement function to obtain an enhanced detail image, including: multiplying the contrast enhancement function with the high-frequency image; and restricting the result of the multiplication to obtain the enhanced detail image.

[0016] In one possible implementation, the high dynamic range infrared raw image is subjected to brightness-preserving dual-platform histogram dimming processing to obtain a dimmed image, including: setting an upper limit threshold for the image statistical histogram to obtain the cumulative distribution function of the histogram, wherein the dual-platform histogram is two segmented platform histograms obtained by segmenting the cumulative distribution function of the histogram; and performing non-interfering histogram mapping on the two platform histograms to obtain the dimmed image.

[0017] On the other hand, embodiments of the present invention provide a high dynamic range infrared image detail enhancement system, comprising:

[0018] The denoising module is used to denoise the high dynamic range infrared raw image using guided filtering to obtain a denoised image.

[0019] The layering module is used to perform image layering processing on the denoised image using Gaussian filtering to obtain a low-frequency image, and to obtain a high-frequency image using the low-frequency image and the high dynamic range infrared original image.

[0020] The enhancement module is used to obtain a contrast enhancement function based on the characteristics of human visual perception, and to use the contrast enhancement function to enhance the high-frequency image to obtain an enhanced detail image.

[0021] The dimming module is used to perform brightness-preserving dual-platform histogram dimming processing on the original high dynamic infrared image to obtain the dimmed image.

[0022] The fusion module is used to perform weighted fusion of the dimmed image and the enhanced detail image.

[0023] On the other hand, embodiments of the present invention provide a computer storage medium storing a plurality of computer instructions, which are used to cause a computer to execute the above-described method.

[0024] The high dynamic range infrared image detail enhancement method, system, and computer storage medium of this invention have the following advantages:

[0025] (1) The present invention uses guided filtering to denoise the original image before layering, and combines Gaussian filtering to directly extract the high-frequency image after noise suppression. The image data contains complete edge information.

[0026] (2) In high-frequency images, this invention combines the characteristics of human visual perception to achieve adaptive selection of enhanced edge weights, thereby improving the intensity of weaker details while protecting the original clear edges.

[0027] (3) This invention directly uses the original image for dual-platform histogram equalization dimming, avoiding edge diffusion caused by filtering algorithms on the original image. The weighted fusion of the dimmed image and the enhanced high-frequency image can effectively reduce the false edge phenomenon. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 A flowchart of a high dynamic range infrared image detail enhancement method provided in an embodiment of the present invention.

[0030] Figure 2 Figure A shows a comparison of the effects of a high dynamic range infrared image detail enhancement method provided in an embodiment of the present invention.

[0031] Figure 3 Figure B shows a comparison of the effects of a high dynamic range infrared image detail enhancement method provided in an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] After researching existing technologies, the applicant discovered that in 2009, Branditta and Francesco proposed a dynamic range segmentation algorithm based on bilateral filtering. Building upon this, scholars from various countries have proposed similar layered filtering framework algorithms. In the bilateral filtering-based dynamic range segmentation algorithm, the bilateral filter divides the original image into a base layer component containing low-frequency information and a detail layer component containing high-frequency information. Compression algorithms and noise suppression are then applied to each layer separately, and a suitable fusion ratio is selected to fuse the images of the two layers. This layered algorithm enhances the contrast of the infrared image while preserving its detail. However, bilateral filters are prone to gradient flipping at edges where image grayscale changes are significant, resulting in halo artifacts and false edges. To eliminate gradient flipping and reduce the overall algorithm computation time, Liu et al. proposed a guided filtering-based DDE algorithm in 2014. The DDE algorithm uses guided filtering instead of bilateral filtering, preserving image detail while avoiding gradient flipping. Because the algorithm's parameter-based approach results in poor scene adaptability, Zhou et al. proposed an adaptive infrared image detail enhancement algorithm based on guided filtering in 2018 to achieve adaptive scene parameter adjustment. This algorithm uses histogram distribution information to determine an adaptive threshold for the base layer image, removing invalid grayscale values ​​and allowing the base layer image to better display valid information. However, the algorithm's denoising effect is poor; when the image contains a large area of ​​sky background, the processed image contains significant noise interference.

[0034] To address the problems in existing technologies, this invention provides a method for enhancing details in high dynamic range infrared images. By first denoising the original high dynamic range infrared image and then performing layered enhancement processing on the denoised image, the problem of severe noise interference in the processed image is solved. By combining the characteristics of human visual perception, the enhancement edge weights are adaptively selected, achieving the effect of improving the intensity of weaker details while protecting the original clear edges. By weightedly fusing the dimmed image and the enhanced image, the false edge phenomenon that occurs during image processing is reduced.

[0035] Figure 1 A flowchart illustrating a high dynamic range infrared image detail enhancement method provided in this embodiment of the invention. This embodiment of the invention provides a high dynamic range infrared image detail enhancement method, comprising:

[0036] A denoised image is obtained by using guided filtering to denoise the original high dynamic range infrared image.

[0037] The denoised image is processed by Gaussian filtering to obtain a low-frequency image, and a high-frequency image is obtained by using the low-frequency image and the original high-dynamic infrared image.

[0038] The contrast enhancement function is obtained based on the visual characteristics of the human eye;

[0039] The high-frequency image is enhanced using the contrast enhancement function to obtain an enhanced detail image.

[0040] The high dynamic range infrared raw image is subjected to brightness-preserving dual-platform histogram dimming processing to obtain a dimmed image;

[0041] The dimmed image and the enhanced detail image are then weighted and fused.

[0042] For example, denoising a high dynamic range infrared (HMR) original image using guided filtering to obtain a denoised image includes: determining the weight coefficients and bias coefficients of the guided filtering based on the HMR original image and the guided image; and filtering the HMR original image based on the guided image, the weight coefficients, and the bias coefficients to obtain the denoised image.

[0043] The guided filtering is a representative edge-preserving smoothing technique. Assume there is a linear relationship between the denoised image q and the guided image I, i.e.: Where, q i I represents the grayscale value of the denoised image at coordinate i. i ω represents the grayscale value of the guide image at coordinate i. k This represents a rectangular window centered at coordinate k with radius r; a k b k These represent the weighting and biasing coefficients of the guiding image within the window, respectively. Guided filtering aims to find the value of a that minimizes the difference between the original high dynamic range infrared image p and the processed denoised image q. k With b k The optimal solution is found; therefore, a cost function needs to be constructed: E(a) k ,b k ), that is E(a k ,b k )=∑((a k I i +b k -p i ) 2 +εa k 2 ) in E(a k ,b k In the formula, ε represents the regularization coefficient, which is used to delete excessively large a values. k And through calculation, we can obtain:

[0044]

[0045]

[0046] Where |ω| represents the number of pixels within the window, σ 2 With μ k These represent the guiding image I at ω. k Variance and mean within the window. This represents the high dynamic range infrared raw image p at ω. k The mean within the window. Finally, a window operation is performed on the entire image, and the mean is taken to obtain the filtered result: in,

[0047] The process involves performing image layering processing on the denoised image using Gaussian filtering to obtain a low-frequency image, and then using the low-frequency image and the original high-dynamic infrared image to obtain a high-frequency image. This includes: setting a Gaussian filter template mask; performing a convolution operation between the Gaussian filter template mask and the original high-dynamic infrared image to obtain the low-frequency image; and subtracting the original high-dynamic infrared image from the low-frequency image to obtain the high-frequency image.

[0048] The Gaussian filter template mask is shown below as a 5*5 Gaussian filter template with a variance of 0.5.

[0049]

[0050]

[0051] The Gaussian filter template mask is used to perform a convolution operation with stride 1 on the denoised image q to obtain the low-frequency image q_L. Then, the difference between q and q_L is used to obtain the high-frequency image q_H, that is: q_H=q-q_L.

[0052] The contrast enhancement function is composed of the product of a local detail adjustment function and a contrast limiting function.

[0053] Let the contrast enhancement function be β(x,y). A local detail adjustment function is constructed using a Weber curve simulation to enhance details based on the background illumination component. This local detail adjustment function is β1(x,y) = 1 - 0.85·sin(q_L'(x,y)·π). Considering practical applications, areas with large gradients in the original high dynamic range infrared (HVR) image do not require excessive enhancement; the real emphasis is placed on areas with smaller gradients. Therefore, a contrast limiting function k(x,y) is constructed. The product of the detail adjustment function β1(x,y) and the contrast limiting function k(x,y) is the final contrast enhancement function β(x,y), i.e., β(x,y) = β1(x,y) × k(x,y). Employing a local contrast enhancement function that simulates human visual characteristics and suppresses excessive detail enhancement, this approach minimizes enhancement in high-contrast areas, resulting in a smoother overall image enhancement without excessive sharpening. It provides strong enhancement for pixels in both dark and bright areas with indistinct details, while simultaneously suppressing over-enhancement of details. In the above formula, q_L'(x,y) and q_H'(x,y) are the normalized variables of q_L(x,y) and q_H(x,y), respectively.

[0054] The process of enhancing the high-frequency image using the contrast enhancement function to obtain an enhanced detail image includes: multiplying the contrast enhancement function with the high-frequency image; and restricting the result of the product to obtain the enhanced detail image.

[0055] The contrast enhancement function β(x,y) is multiplied with the high-frequency image q_H(x,y), and the product result is restricted to the range [-20, 20] to obtain q_HE(x,y) as the enhanced detail image.

[0056] In one possible embodiment, the high dynamic range infrared raw image is subjected to brightness-preserving dual-platform histogram dimming processing to obtain a dimmed image, including: setting an upper limit threshold for the image statistical histogram to obtain a cumulative histogram distribution function, wherein the dual-platform histogram is two segmented platform histograms obtained by segmenting the cumulative histogram distribution function; and performing non-interfering histogram mapping on the two platform histograms to obtain the dimmed image.

[0057] Set the image statistical histogram T as the upper threshold. When the gray-level frequency P is greater than T, the value of T is assigned to P; otherwise, P remains unchanged. Here, T = 500 is taken. The cumulative distribution function of the histogram can be expressed as follows: Using 8192 as the center point of the histogram segment, the first 8192 gray levels are mapped to [y8Start, y8Mid], and the next 8192 gray levels are mapped to [y8Mid, 255]. y8Range represents the dynamic range of an 8-bit image; here, y8Range = 235. The dimmed image q_base(x,y) can be obtained by performing independent histogram mapping on the two platform histograms.

[0058] The dimmed image q_base(x,y) and the enhanced high-frequency image q_HE(x,y) are weighted and fused. That is: oImg(x,y) = q_base(x,y) + q_HE(x,y); the resulting oImg(x,y) is the final enhanced infrared image, such as... Figure 2 and Figure 3 As shown. Among them, Figure 2 In the image, 'a' is the high dynamic range infrared image before detail enhancement, and 'b' is the high dynamic range infrared image after detail enhancement. Figure 3 In the image, 'a' is the high dynamic range infrared image before detail enhancement, and 'b' is the high dynamic range infrared image after detail enhancement.

[0059] This invention also provides a high dynamic range infrared image detail enhancement system, comprising:

[0060] The denoising module is used to denoise the high dynamic range infrared raw image using guided filtering to obtain a denoised image.

[0061] The layering module is used to perform image layering processing on the denoised image using Gaussian filtering to obtain a low-frequency image, and to obtain a high-frequency image using the low-frequency image and the high dynamic range infrared original image.

[0062] The enhancement module is used to obtain a contrast enhancement function based on the characteristics of human visual perception, and to use the contrast enhancement function to enhance the high-frequency image to obtain an enhanced detail image.

[0063] The dimming module is used to perform brightness-preserving dual-platform histogram dimming processing on the original high dynamic infrared image to obtain the dimmed image.

[0064] The fusion module is used to perform weighted fusion of the dimmed image and the enhanced detail image.

[0065] This invention also provides a computer storage medium storing a plurality of computer instructions, which are used to cause a computer to execute the above-described method.

[0066] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0067] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for enhancing details in high dynamic range infrared images, characterized in that, include: A denoised image is obtained by using guided filtering to denoise the original high dynamic range infrared image. The denoised image is processed by Gaussian filtering to obtain a low-frequency image, and a high-frequency image is obtained by using the low-frequency image and the original high-dynamic infrared image. The contrast enhancement function is obtained based on the visual characteristics of the human eye; The high-frequency image is enhanced using the contrast enhancement function to obtain an enhanced detail image. The high dynamic range infrared raw image is subjected to brightness-preserving dual-platform histogram dimming processing to obtain a dimmed image; The dimmed image and the enhanced detail image are then weighted and fused. The process involves performing brightness-preserving dual-platform histogram dimming on the original high-dynamic infrared image to obtain a dimmed image, including: Set an upper limit threshold T for the image statistical histogram. When the grayscale frequency P is greater than T, assign the value of T to P. Otherwise, P remains unchanged. This yields the cumulative distribution function of the histogram. The dual-platform histogram is the two-segment platform histogram obtained by segmenting the cumulative distribution function of the histogram. The two platform histograms are mapped independently to obtain the adjusted image. Specifically, 8192 is used as the center point of the histogram segment. The first 8192 gray levels are mapped to [y8Start, y8Mid], and the last 8192 gray levels are mapped to [y8Mid, 255]. y8Range represents the dynamic range of an 8-bit image.

2. The high dynamic range infrared image detail enhancement method according to claim 1, characterized in that, Denoising of the high dynamic range infrared raw image is performed using guided filtering to obtain a denoised image, including: The weighting and biasing coefficients of the guided filter are determined based on the original high-dynamic infrared image and the guided image. The high dynamic range infrared original image is filtered based on the guiding image, weighting coefficients, and bias coefficients to obtain the denoised image.

3. The high dynamic range infrared image detail enhancement method according to claim 1, characterized in that, The denoised image is processed by Gaussian filtering to obtain a low-frequency image, and a high-frequency image is obtained using the low-frequency image and the original high-dynamic infrared image, including: Set a Gaussian filter template mask; The low-frequency image is obtained by convolving the Gaussian filter template mask with the high dynamic range infrared original image; The high-frequency image is obtained by subtracting the high-dynamic infrared original image from the low-frequency image.

4. The high dynamic range infrared image detail enhancement method according to claim 1, characterized in that, The contrast enhancement function is composed of the product of a local detail adjustment function and a contrast limiting function.

5. The high dynamic range infrared image detail enhancement method according to claim 1, characterized in that, The high-frequency image is enhanced using the contrast enhancement function to obtain an enhanced detail image, including: Multiply the contrast enhancement function with the high-frequency image; The result of the product is constrained to obtain the enhanced detail image.

6. A high dynamic range infrared image detail enhancement system, wherein the system applies the high dynamic range infrared image detail enhancement method according to any one of claims 1-5, characterized in that, include: The denoising module is used to denoise the high dynamic range infrared raw image using guided filtering to obtain a denoised image. The layering module is used to perform image layering processing on the denoised image using Gaussian filtering to obtain a low-frequency image, and to obtain a high-frequency image using the low-frequency image and the high dynamic range infrared original image. The enhancement module is used to obtain a contrast enhancement function based on the characteristics of human visual perception, and to use the contrast enhancement function to enhance the high-frequency image to obtain an enhanced detail image. The dimming module is used to perform brightness-preserving dual-platform histogram dimming processing on the original high dynamic infrared image to obtain the dimmed image. The fusion module is used to perform weighted fusion of the dimmed image and the enhanced detail image.

7. A computer storage medium, characterized in that, The computer storage medium stores a plurality of computer instructions, which are used to cause the computer to perform the method described in any one of claims 1-5.

Images