A diffraction image restoration method based on block minimum pixel prior
Through the method based on the block minimum pixel prior and combined with the diffraction image degradation model, the diffraction image blur and chromatic aberration problems are solved, and more efficient image restoration effect is achieved, significantly improving the sharpness and quality of the image.
Patent Information
- Application Number
- CN202210615504.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-01
- Publication Date
- 2025-04-08
- Estimated Expiration
- 2042-06-01
AI Technical Summary
Due to blur and chromatic aberration problems of diffraction images, the existing restoration methods are limited in effect and are difficult to meet the needs of high resolution.
The diffraction image restoration method based on the block minimum pixel prior is adopted, combined with the diffraction image degradation model, by initializing the blurred core and calculating the block minimum pixel prior, using L0 norm constraints and L2 norm optimization, the blurred core and clear images iteratively update.
It significantly improves the clarity of the diffraction image, reduces the chromatic aberration problem, improves the peak signal-to-noise ratio and grayscale average gradient, and has a more significant effect than the original method.
Smart Images
Figure CN115018725B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of image processing and optical engineering technology, specifically to the field of diffraction image restoration, and in particular to a diffraction image restoration method based on block minimum pixel prior. Background Art
[0002] With the advancement of science and technology, modern imaging technology has been able to assist humans in recording, observing and analyzing the world; however, the increase in resolution requirements has also restricted the development of imaging systems. Thin-film diffraction imaging systems have gradually entered the field of vision of scientific researchers due to their advantages of light weight and miniaturization. Thin-film diffraction elements solve the problems of heavy weight and high precision of traditional optical systems, thus providing a way to design large-aperture, lightweight optical systems.
[0003] Unlike traditional optical systems, the image formed by a diffraction system has a low contrast and is easily affected by various noises. At the same time, the quality of the image is closely related to the diffraction efficiency of the diffraction elements in the optical system. Compared with traditional optical systems, the chromatic aberration of the image formed by the diffraction system is more serious. When processing images, not only the blur problem but also the chromatic aberration problem needs to be considered. At present, the restoration problem of diffraction images is divided into two categories. Like traditional images, deep learning methods and traditional image restoration methods are used. However, due to the complex degradation process of the diffraction system, the restoration method of the diffraction image is limited. Summary of the invention
[0004] Aiming at the blur and chromatic aberration problems of diffraction images, the present invention proposes a diffraction image restoration method based on block minimum pixel prior. In the analysis of the imaging process of the system, it is found that the minimum value of the image pixel will increase with the image blurring process, resulting in image degradation. Therefore, based on this prior knowledge and combined with the diffraction image degradation model, an image restoration method is designed.
[0005] The technical solution adopted by the method of the present invention is: a diffraction image restoration method based on block minimum pixel prior, comprising the following steps:
[0006] Step 1: Input a blurred diffraction image formed by a diffraction optical system.
[0007] Step 2: Establish an objective function model of the blurred diffraction image based on the degradation model of the diffraction optical system.
[0008] Step 3: Initialize the blur kernel of the diffraction image degradation process.
[0009] Step 4: Calculate the block minimum pixel prior of the input blurred diffraction image, constrain it using the L0 norm, and calculate the mask based on the prior.
[0010] Step 5: Convert the problem in step 3 into sub-problems, and use the auxiliary variable G of the gradient of the input diffraction image and iteratively update the blurred diffraction image based on prior knowledge and the mask.
[0011] Step 6: According to the iterative blurred diffraction image, the initialized blur kernel iteratively updates the blur kernel, and finally solves the blur kernel and the clear image, and outputs the final clear image, which is the same size as the input blurred diffraction image.
[0012] Furthermore, the objective function model of step 2 is:
[0013]
[0014] Where K = η int k-(1-η int )δ(x,y),η int represents the diffraction efficiency of the diffraction system, and k represents the blur kernel of the diffraction image. The above formula is constrained by p(x), which is the probability distribution function of the minimum pixel prior of the input blurred diffraction image. In the first term In the equation, I represents the estimated clear image, and B represents the blurred image. represents the convolution operation, χ is the credibility parameter, which is constrained by the L2 norm to minimize the loss of the value after the convolution of the clear image and the blur kernel and the blur image; the second term is to constrain the blur kernel using the L2 norm, where γ is the regularization parameter; the third term μ is the regularization parameter, Represents the image gradient matrix, which is constrained by the L0 norm to remove local small gradient areas of the image and retain large gradient areas.
[0015] Furthermore, the blur kernel refers to the function that causes the system image to degrade, which is expressed as a two-dimensional matrix. The blur kernel initialization process in step 3 is: according to the blur kernel size m set by the input, create an m*m matrix with all values of 0, and then set the element value of the row position (m-1 / 2) and the column position (m-1 / 2) to 0.1.
[0016] Furthermore, the block (local) minimum pixel refers to dividing the input image into blocks and finding the minimum pixel in each block. The calculation method is as follows:
[0017]
[0018] Among them, Ω i represents the i-th image block, and (x, y, c) represents the pixel value of any point in the image.
[0019] Furthermore, the sub-problems into which the objective function is decomposed in step 5 are as follows:
[0020] Formula 1:
[0021] Formula 2:
[0022] Where K = η int k+(1-η int )δ(x,y), G is the auxiliary variable of the image gradient of the input blurred diffraction image, Formula 1 is used to iteratively solve the clear image, and Formula 2 is used to iteratively solve the blur kernel.
[0023] The advantages of the present invention compared with the prior art are:
[0024] (1) The present invention combines the image restoration method based on block minimum pixel prior with the diffraction imaging degradation model, and optimizes the problem that the original method has less than ideal diffraction image restoration effect. Compared with the original method, the restored image has smaller chromatic aberration problem of the diffraction image.
[0025] (2) The present invention uses an image restoration method based on block minimum pixel prior, which is more computationally efficient than other traditional methods.
[0026] (3) The present invention combines the diffraction image degradation model. Compared with the original method, the peak signal-to-noise ratio, grayscale average gradient, etc. are improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 This is a flow chart of a diffraction image restoration method based on block minimum pixel prior according to the present invention.
[0028] Figure 2 This is a schematic diagram of taking pictures for the system.
[0029] Figure 3 This is a schematic diagram of an image processed by the method of the present invention.
[0030] Figure 4 This is a schematic diagram of the second picture taken for the system.
[0031] Figure 5 This is a second schematic diagram of an image processed by the method of the present invention. DETAILED DESCRIPTION
[0032] The present invention is further described below with reference to the accompanying drawings and specific embodiments.
[0033] like Figure 1 As shown, a diffraction image restoration method based on block minimum pixel prior includes the following steps:
[0034] Step 1: Input the diffraction image matrix. Input a blurred diffraction image formed by a diffraction optical system.
[0035] Step 2: Establish the objective function of blurred diffraction image restoration based on the degradation model of the diffraction optical system. The objective function is as follows:
[0036]
[0037] Where K = η int k-(1-η int )δ(x,y),η int represents the diffraction efficiency of the diffraction system, and k represents the blur kernel of the diffraction image. The above formula is constrained by p(x), which is the probability distribution function of the minimum pixel of the input blurred diffraction image. In the equation, I represents the estimated clear image, and B represents the blurred image. represents the convolution operation, χ is the credibility parameter, which is constrained by the L2 norm to minimize the loss of the value after the convolution of the clear image and the blur kernel and the blur image; the second term is to constrain the blur kernel using the L2 norm, where γ is the regularization parameter; the third term μ is the regularization parameter, Represents the image gradient matrix, which is constrained by the L0 norm to remove local small gradient areas of the image and retain large gradient areas.
[0038] Step 3: Initialize the blur kernel of the system imaging process. The blur kernel refers to the function that causes the diffraction image to degrade, which is a two-dimensional matrix of size m*m. The process of initializing the blur kernel is: according to the blur kernel size m set by the input, create an m*m matrix with all values of 0, and then set the element values of the row position (m-1 / 2) and the column position (m-1 / 2) to 0.1.
[0039] Step 4: Calculate the block minimum pixel prior of the image and use it as the L0 norm for sparse constraint. Block minimum pixel means dividing the input image into blocks and calculating the minimum pixel value in each image block. The calculation method is as follows:
[0040]
[0041] Among them, Ω i represents the i-th image block, and (x, y, c) represents the pixel value of any point in the image.
[0042] Step 5: Decompose the objective function in step 3 into sub-problems. The sub-problems decomposed into the objective function are as follows:
[0043] Formula 1:
[0044] Formula 2:
[0045] Where K = η int k-(1-η int )δ(x,y), G is the auxiliary variable of the image gradient of the input blurred diffraction image, Formula 1 is used to solve the clear image I, and Formula 2 is used to solve the blur kernel k.
[0046] Step 6: Iteratively solve the estimated blur kernel and clear image using the semi-quadratic splitting method.
[0047] Step 7: Output the final blur kernel with a size of m*m. Output the final clear image with the same size as the input image.
[0048] Compared with the diffraction image obtained by direct shooting, the clarity of the diffraction image processed by the present invention is greatly improved. Figure 2 , Figure 4 They are system-captured picture 1, system-captured picture 2, and Figure 3 , Figure 5 The following are schematic diagrams of images processed by this method. According to the measurement, the peak signal-to-noise ratio of the image processed by this method is increased by 2.18dB to 4.43dB on average, and the average grayscale gradient is increased by 0.314 to 0.632.
[0049] It should be understood that the above description of the preferred embodiment is relatively detailed and cannot be regarded as limiting the scope of patent protection of the present invention. Under the enlightenment of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested for the present invention shall be based on the attached claims.
Claims
1. A diffraction image restoration method based on block minimum pixel prior, characterized by: The following steps are involved: Step 1: Input a blurred diffraction image formed by a diffraction optical system; Step 2: Establish an objective function model of the blurred diffraction image according to the degradation model of the visible light diffraction imaging system; the objective function model is: in, , η int represents the diffraction efficiency of the diffraction system, k represents the blur kernel of the diffraction image, and the above formula is constrained by p(x), which is the probability distribution function of the block minimum pixel prior of the input blurred diffraction image. In the equation, I represents the estimated clear image, and B represents the blurred image. represents the convolution operation, is the credibility parameter, which is constrained by the L2 norm to minimize the loss of the value after the convolution of the clear image and the blur kernel and the blur image; the second term The L2 norm is used to constrain the blur kernel, where is the regularization parameter; the third term , μ is the regularization parameter, Represents the image gradient matrix, which is constrained by the L0 norm to remove local small gradient areas of the image and retain large gradient areas; Step 3: Initialize the blur kernel of the blurred diffraction image degradation process; Step 4: Calculate the block minimum pixel prior of the input blurred diffraction image, constrain it using the L0 norm, and calculate the mask based on the prior; the block minimum pixel refers to dividing the input image into blocks and finding the minimum pixel in each block. The calculation method is as follows: Among them, Ω i represents the i-th image block, (x, y, c) represents the pixel value of any point in the image; Step 5: Convert the problem in step 3 into sub-problems, and use the auxiliary variable G of the gradient of the input diffraction image and iteratively update the blurred diffraction image based on prior knowledge and the mask; Step 6: According to the iterative blurred diffraction image, the initialized blur kernel iteratively updates the blur kernel, and finally solves the blur kernel and the clear image, and outputs the final clear image, which is the same size as the input blurred diffraction image.
2. A diffraction image restoration method based on block minimum pixel prior according to claim 1, characterized in that: The blur kernel refers to the function that causes the system image to degrade, which is expressed as a two-dimensional matrix. The blur kernel initialization process in step 3 is: according to the blur kernel size m set by the input, create an m matrix with all values 0 m matrix, and then set the element values at row position (m-1 / 2) and column position (m-1 / 2) to 0.
1.
3. A diffraction image restoration method based on block minimum pixel prior according to claim 1, characterized in that: In step 5, the objective function is decomposed into the following sub-problems: Formula 1: Formula 2: in, , G is the auxiliary variable of the image gradient of the input blurred diffraction. Formula 1 is used to iteratively solve the clear image, and Formula 2 is used to iteratively solve the blur kernel.