A calibration-based adaptive image restoration method

Abstract

The present invention belongs to the field of image restoration processing, and specifically discloses a calibration-based image adaptive restoration method, including the following steps: noise calibration, photographing and saving a set target, and performing noise calibration; imaging system calibration; image denoising; image deblurring; and outputting the final restored image. The present invention calibrates the noise and blur coefficients of an actual optical system, performs parameterized mathematical modeling, and applies the calibration results to a subsequent image restoration algorithm to obtain the best image restoration result for this imaging system, thereby greatly improving the image restoration effect.

Description

Technical Field

[0001] The invention relates to the field of image restoration processing, and in particular to an image adaptive restoration method based on calibration. Background Art

[0002] The image formed by a point in the object space through an ideal optical imaging system is still a point. However, due to the influence of geometric aberration and optical diffraction limit, the image of the actual optical imaging system is no longer a clear and perfect image, but has a certain optical blur. At the same time, the imaging process is always accompanied by noise signals, which mainly come from Poisson noise related to light intensity, transfer noise introduced in the charge transfer process, dark current noise generated by thermal excitation on the silicon substrate, and non-uniformity noise.

[0003] These factors lead to the degradation of image quality, so image denoising and deblurring are very important steps in image restoration. Image denoising generally starts from the smoothness of the image, the high frequency of the frequency domain and the randomness of the distribution, and constructs a corresponding mathematical model to remove noise; image deblurring uses human cognition of clear images to add priors to the degradation model for model solution to achieve the function of image deblurring.

[0004] Currently, most algorithms for image processing artificially make certain assumptions, such as noise obeying a Gaussian distribution or a blur kernel default value, which results in unsatisfactory image restoration effects for different imaging systems. Summary of the invention

[0005] The purpose of the present invention is to provide a calibration-based image adaptive restoration method to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a calibration-based image adaptive restoration method, comprising the steps of:

[0007] S1: Noise calibration, photographing and saving the set target, and performing noise calibration;

[0008] S2: Imaging system calibration;

[0009] S3: Image denoising processing;

[0010] S4: image deblurring;

[0011] S5: Output the final restored image.

[0012] Preferably, the noise calibration in S1 includes global noise calibration and local noise calibration, and the specific steps of noise calibration include:

[0013] S101: setting a target, photographing the target using an actual optical system, and saving the photographed image;

[0014] S102: Select two target images of the same scene at the same time;

[0015] S103: Calculate global image noise distribution;

[0016] S104: Calculate local image noise distribution.

[0017] Preferably, the global image noise distribution calculation process in S103 specifically includes:

[0018] S103a: performing a difference operation on two frames of target images to obtain a difference image;

[0019] S103b: performing distribution fitting of a probability density function on the difference image to determine its distribution type and specific parameter values;

[0020] S103c: Save the result into a parameter file.

[0021] Preferably, the local image noise distribution calculation process in S104 specifically includes:

[0022] S104a: intercepting N image blocks with different gray values ​​on the target image, where N is 8 by default;

[0023] S104b: Draw a histogram for each image block;

[0024] S104c: performing distribution fitting of the probability density function to determine its distribution type and specific parameter values;

[0025] S104d: average the results of N image blocks;

[0026] S104e: Save all results into a parameter file.

[0027] Preferably, the imaging system calibration process of S2 specifically includes:

[0028] S201: Select a frame of target image;

[0029] S202: intercepting the region ROI with black and white oblique edges in the image;

[0030] S203: sequentially "project" data of different rows in the ROI onto the same pixel grid to obtain an edge spread function ESF;

[0031] S204: Deriving the edge spread function ESF to obtain a rate of change line spread function LSF;

[0032] S205: Perform Fourier transform FFT on LSF to obtain the response value SFR at each spatial frequency;

[0033] S206: Analyze the SFR data to obtain the imaging system blur parameter;

[0034] S207: Store the result in a parameter file.

[0035] Preferably, the image denoising process in S3 is implemented based on a distribution algorithm, and the distribution algorithm includes but is not limited to a median filter, a non-local means algorithm, and an RL algorithm.

[0036] Preferably, the image deblurring process in S4 is implemented based on deblurring calculation, and the image after deblurring processing is the final restored image.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] The present invention calibrates the noise distribution and imaging system through the current optical system, saves the calibrated file, applies it to the subsequent algorithm as a priori parameter, and selects different algorithms according to the noise distribution to achieve the best denoising effect; the result of the imaging system calibration is used as the initial value of the edge deblurring algorithm to further improve the correctness of the blur kernel estimation of blind deblurring, thereby greatly improving the deblurring effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flowchart of the whole process of the present invention;

[0040] Figure 2 A flowchart of noise calibration in an embodiment of the present invention;

[0041] Figure 3 A flowchart of the calibration of the imaging system in an embodiment of the present invention;

[0042] Figure 4 A flowchart of image denoising in an embodiment of the present invention;

[0043] Figure 5 is a flowchart of image deblurring according to an embodiment of the present invention;

[0044] Figure 6 Schematic diagram of target calibration in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0046] See also Figure 1The present invention provides a technical solution: a calibration-based image adaptive restoration method, comprising the steps of:

[0047] S1: Noise calibration, take pictures of the set target, save it, and perform noise calibration (calibration target such as Figure 6 );

[0048] S2: Imaging system calibration;

[0049] S3: Image denoising processing;

[0050] S4: image deblurring;

[0051] S5: Output the final restored image.

[0052] See also Figure 2 In this embodiment, the noise calibration in S1 includes global noise calibration and local noise calibration. The specific steps of noise calibration include:

[0053] S101: setting a target, photographing the target using an actual optical system, and saving the photographed image;

[0054] S102: Select two target images of the same scene at the same time;

[0055] S103: Calculate global image noise distribution;

[0056] S104: Calculate local image noise distribution.

[0057] In this embodiment, the global image noise distribution calculation process in S103 specifically includes:

[0058] S103a: performing a difference operation on two frames of target images to obtain a difference image;

[0059] S103b: performing distribution fitting of a probability density function on the difference image to determine its distribution type and specific parameter values;

[0060] S103c: Save the result into a parameter file.

[0061] In this embodiment, the local image noise distribution calculation process in S104 specifically includes:

[0062] S104a: intercepting N image blocks with different gray values ​​on the target image, where N is 8 by default;

[0063] S104b: Draw a histogram for each image block;

[0064] S104c: performing distribution fitting of the probability density function to determine its distribution type and specific parameter values;

[0065] S104d: average the results of N image blocks;

[0066] S104e: Save all results into a parameter file.

[0067] See also Figure 3 In this embodiment, the imaging system calibration process of S2 specifically includes:

[0068] S201: Select a frame of target image;

[0069] S202: intercepting the region ROI with black and white oblique edges in the image;

[0070] S203: sequentially "project" data of different rows in the ROI onto the same pixel grid to obtain an edge spread function ESF;

[0071] S204: Deriving the edge spread function ESF to obtain a rate of change line spread function LSF;

[0072] S205: Perform Fourier transform FFT on LSF to obtain the response value SFR at each spatial frequency;

[0073] S206: Analyze the SFR data to obtain the imaging system blur parameter; S207: Store the result in a parameter file.

[0074] See also Figure 4 In this embodiment, the image denoising process in S3 is implemented based on a distribution algorithm, and the distribution algorithm includes but is not limited to a median filter, a non-local mean algorithm, and an RL algorithm.

[0075] S3 specifically includes the following steps:

[0076] S301: Import parameter file;

[0077] S302: Select an input image to be processed;

[0078] S303: Select an adaptive algorithm according to the noise distribution calibrated by the parameter file. The algorithm provides adaptive algorithm selection; the algorithm includes:

[0079] a. Median filtering (salt and pepper noise): Median filtering is a nonlinear signal processing technology based on sorting statistics theory that can effectively suppress noise. The basic principle of median filtering is to replace the value of a point in a digital image or digital sequence with the median value of each point in a neighborhood of the point, thereby eliminating isolated noise points;

[0080] b. Non-local mean algorithm (Gaussian noise). The non-local mean algorithm is a method of smoothing and averaging the area around a target pixel, so non-local mean filtering means that it uses all pixels in the image, and these pixels are weighted averaged according to a certain similarity. The image after filtering is clear and does not lose details; the non-local mean algorithm uses the redundant information commonly found in natural images to remove noise. Unlike bilinear filtering, median filtering, etc., which use local information of the image to filter, it uses the entire image for denoising, that is, it searches for similar areas in the image in units of image blocks, and then averages these areas to better filter out Gaussian noise in the image;

[0081] c. RL algorithm (Poisson noise);

[0082] S304: Processing the input image;

[0083] S305: Output the processed image and then enter the image deblurring process.

[0084] See also Figure 5 In this embodiment, the image deblurring process in S4 is implemented based on deblurring calculation, and the image after deblurring processing is the final restored image.

[0085] In this embodiment, S4 specifically includes:

[0086] S401: Read the imaging system blur parameters calibrated by the parameter file and construct a two-dimensional kernel function;

[0087] S402: using the kernel obtained in S401 as the initial value of the edge deblurring algorithm to start image deblurring;

[0088] S403: After the image is deblurred, the restoration operation is completed.

[0089] The above embodiment calibrates the noise and blur coefficients of the actual optical system, performs parameterized mathematical modeling, and applies the calibration results to the subsequent image restoration algorithm to obtain the best image restoration result for the imaging system.

[0090] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A calibration-based image adaptive restoration method, characterized in that: Includes steps: S1: Noise calibration, photograph and save the set target, and perform noise calibration; The noise calibration in S1 includes global noise calibration and local noise calibration. The specific steps of noise calibration include: S101: setting a target, photographing the target using an actual optical system, and saving the photographed image; S102: Select two target images of the same scene at the same time; S103: Calculate global image noise distribution; The global image noise distribution calculation process in S103 specifically includes: S103a: performing a differential operation on two frames of target images to obtain a difference image; S103b: performing a distribution fitting of a probability density function on the difference image to determine its distribution type and specific parameter values; S103c: storing the result in a parameter file; S104: Calculate local image noise distribution; The local image noise distribution calculation process in S104 specifically includes: S104a: intercepting N image blocks with different gray values ​​on the target image, where N is 8 by default; S104b: drawing a histogram for each image block; S104c: performing distribution fitting of a probability density function to determine its distribution type and specific parameter value; S104d: averaging the results of the N image blocks; S104e: storing all the results in a parameter file; S2: Imaging system calibration; The S2 imaging system calibration process specifically includes: S201: Select a frame of target image; S202: intercepting the region ROI with black and white oblique edges in the image; S203: sequentially "project" data of different rows in the ROI onto the same pixel grid to obtain the edge spread function ESF; S204: Deriving the edge spread function ESF to obtain a rate of change line spread function LSF; S205: Perform Fourier transform FFT on LSF to obtain the response value SFR at each spatial frequency; S206: Analyze the SFR data to obtain the imaging system blur parameter; S207: storing the result in a parameter file; S3: Image denoising processing; Image denoising in S3 is implemented based on distribution algorithms, including but not limited to value filtering, non-local mean algorithm, and RL algorithm; S3 specifically includes the following steps: S301: Import parameter file; S302: Select an input image to be processed; S303: Select an adaptive algorithm according to the noise distribution calibrated by the parameter file. The algorithm provides adaptive algorithm selection. The algorithms include: a. median filtering; b. non-local mean algorithm; c. RL algorithm; S304: Processing the input image; S305: output the processed image and then enter the image deblurring process; S4: image deblurring; The image deblurring process in S4 is implemented based on the deblurring calculation. The image after deblurring is the final restored image. S4 specifically includes: S401: Read the imaging system blur parameters calibrated by the parameter file and construct a two-dimensional kernel function; S402: using the kernel obtained in S401 as the initial value of the edge deblurring algorithm to start image deblurring; S403: completing the restoration operation after the image is deblurred; S5: Output the final restored image.

Images