Color scanning image gray level improvement de-mesh super-division reconstruction method

Abstract

According to the color scanning image gray level improved de-netting super-division reconstruction method, de-netting processing is carried out on a scanning image by adopting different wavelet differentiation levels and different weak wave models, processing is carried out by adopting first-level wavelet differentiation and a db4 weak wave model, the de-netting effect is optimal, and then estimation of reticulate noise sigma and selection of a wavelet critical domain T are improved. And primary wavelet differentiation and a db4 weak wave model are introduced into an improved algorithm, image super-resolution reconstruction is carried out on the de-screened image, and image enhancement processing is further carried out on the de-screened image. The method for improving the wavelet critical domain through net removal comprises the steps of establishing a wavelet differentiation series and a weak wave model, estimating the cobwebbing noise variance in wavelet critical domain filtering, calculating the size of the critical domain and evaluating net removal effectiveness, and accurate and efficient cobwebbing removal of a digital image obtained through scanning is achieved. According to the invention, the digital image obtained in the digital processing process of the printed image can be ensured to be capable of accurately and quickly transmitting information.

Description

Technical Field

[0001] The present application relates to a method for removing moiré and super-resolution reconstruction of scanned images, and particularly to a method for improving gray levels of color scanned images for moiré removal and super-resolution reconstruction, belonging to the technical field of moiré removal and reconstruction of scanned images. Background Art

[0002] In daily life and the scientific research field, more digital images are mainly obtained by means of devices such as scanners, digital cameras, mobile phones, printing and fax machines, and computers. However, due to the loss of paper images, people need to rely on image input devices such as scanners to obtain the digital information of paper images again. In recent years, although digital imaging devices have greatly improved in terms of the quality of image acquisition, since paper images adopt discontinuous digital gray level processing in the printing output stage, and information acquisition devices such as scanners also sample at discontinuous equal intervals, the quality of digital images obtained by using devices such as scanners fails to meet the requirements, mainly manifested in that there are some noises in the obtained digital images and regular moiré phenomena appear in the digital images, seriously affecting the visual system of the human eye to read the obtained digital images. When the digital images obtained by using a scanner need to be used for display or other purposes, the obtained digital images do not meet the application requirements, and it is necessary to restore the digital images to ideal continuous tone images. Therefore, moiré removal of color gray level scanned images plays an important role.

[0003] With the continuous development of the "Internet + printing" technology, the digitization of paper document materials and digital publishing have become hot topics. The application of various mobile intelligent terminal communication devices has raised higher requirements for the channels of information acquisition, and real-time and accuracy are the main features. Therefore, based on this demand, the requirements and standards for digital processing of paper printed materials are also getting higher and higher. Existing paper printed images all use digital gray level technology and then are realized through printing. In the process of digitizing these printed images, if the processing is unreasonable, obvious image distortion and image degradation problems will occur in the obtained digital images, which will lead to inaccurate acquisition of digital image information. In order to ensure that the digital images obtained during the digital processing of printed images can accurately and quickly transmit information, it is necessary to perform moiré removal processing on the scanned digital images.

[0004] While the application of digital publishing and digital media information technology is constantly expanding, various publishers have added various copyright anti-counterfeiting means to existing printed products, posing new requirements for anti-counterfeiting technology in printing production materials and the entire printing process. Whether the moiré information of color gray level scanned images is effectively removed is very important for the post-processing of images and the reprinting process. Removing moiré from scanned images not only improves the visual effect of the human eye, but also ensures the stability and invisibility of embedded digital watermark information.

[0005] Problems to be solved in the de-screening super-resolution reconstruction of scanned images in the prior art and key technical difficulties of this application include:

[0006] (1) Since paper images adopt discontinuous digital gray-level processing in the printing output stage, and information acquisition devices such as scanners also sample at discontinuous equal intervals, the quality of digital images obtained by using devices such as scanners does not meet the requirements. There are some noises and regular moiré patterns in the obtained digital images, which seriously affect the visual system of the human eye to read the obtained digital images. When the digital images obtained by using a scanner need to be used for display or other purposes, the obtained digital images do not meet the application requirements, and it is necessary to restore the digital images to ideal continuous-tone images again. The prior art lacks a method for removing moiré patterns from high-quality color gray-level scanned images, lacks a technical solution for improving the de-screening of the wavelet critical region, and has not established the estimation of the wavelet differentiation level and the weak wave model, the variance of moiré noise in wavelet critical region filtering, the calculation of the critical region size, and the evaluation of de-screening effectiveness, and cannot guarantee the data volume and clarity of the image after de-screening.

[0007] (2) Paper printed images in the prior art all use digital gray-level technology and then are realized by printing. During the digitalization process of these printed images, if the processing is unreasonable, the obtained digital images will have obvious image distortion and image degradation problems, which will lead to inaccurate acquisition of digital image information. To ensure that the digital images obtained during the digitalization process of printed images can accurately and quickly transmit information, it is necessary to perform moiré removal processing on the scanned digital images. However, through the analysis of the de-screening method of scanned images, it is found that when using the existing wavelet critical region method to de-screen color gray-level scanned images, a large amount of detail information of the scanned images is lost and the images become blurred. There is a lack of re-determination of the noise variance of the moiré image and the wavelet critical region, and a lack of super-resolution reconstruction processing for the image after de-screening.

[0008] (3) The existing spatial domain method for moire removal mainly processes in a two-dimensional space, and the effect of removing noise with large gray-scale differences is not very good, such as salt-and-pepper noise. While the non-linear median filtering denoising method reduces the noise in digital images, it also causes the phenomenon of blurred digital image edges and serious loss of image details to a certain extent. The quality of the smooth area of the image after moire removal is not high, and the moire information in the gray-level edge area cannot be effectively removed. The existing method for removing moire in the scanned image in the transform domain can achieve relatively good results, but it cannot process the moire in the image in real time and has deficiencies in maintaining texture recognition. The computational complexity and workload of this method are also very large. The algorithm that combines the spatial domain and the frequency domain to remove moire from scanned images is also used more frequently. However, other image noises are introduced during the moire removal process of this algorithm, making it difficult to achieve the requirements of image quality for moire removal, and it also does not meet the requirements of real-time moire processing. Summary of the Invention

[0009] This application analyzes the causes of moire formation in scanned images and designs the moire removal process for color gray-scale scanned images, including the determination of wavelet decomposition levels and weak wave models, the improvement of the method for wavelet critical regions, and the moire removal process for wavelet critical regions. By analyzing the method for selecting critical regions in the existing wavelet critical region moire removal algorithms, the deficiencies are improved, including the estimation of moire noise in scanned images and the selection of wavelet critical regions. The scanned image is decomposed at different scales to obtain image information of different components, and then the improved wavelet critical region is used for filtering processing, so that the scanned image after moire removal can be obtained. The super-resolution reconstruction of the moire-removed image is established to further enhance the detail information of the scanned image. For the evaluation of the moire removal effect of color gray-scale scanned images, two aspects of non-reference texture recognition and scanned data level are selected to evaluate the moire removal effect of the improved algorithm of this application, and it is compared with the moire removal effects of the existing wavelet critical region method, the filter method of the scanner's built-in NewColor7000 software, and the Gaussian blur algorithm of Photoshop, so as to reflect that the moire removal effect of the algorithm of this application is good, and both the data volume and clarity have been greatly improved.

[0010] To achieve the above technical effects, the technical solutions adopted in this application are as follows:

[0011] Color scanning image gray level improvement de-screening super-resolution reconstruction method, which uses different wavelet decomposition levels and different weak wave models to perform de-screening processing on the scanning image. The first-level wavelet decomposition and db4 weak wave model are used for processing, and the de-screening effect is the best. Then, the estimation of the screen noise σ and the selection of the wavelet critical domain T are improved, and the first-level wavelet decomposition and db4 weak wave model are introduced into the improved algorithm. Finally, based on the improved algorithm of this application, after de-screening the scanning image, it will further cause the loss of detail information and the blurring of the image. Image super-resolution reconstruction is performed on the de-screened image, and further image enhancement processing is performed on the de-screened image;

[0012] Based on wavelet critical domain de-screening, first separate the channels of the color gray-level scanning image, and then process each color channel of R, G, and B obtained by separation. The specific method is as follows: According to the screen distribution data of each color channel gray image, perform wavelet decomposition on the gray image of each color channel based on the optimal combination of wavelet decomposition levels and weak wave models, so that the screen is most likely to be distributed on the high-frequency components of the image after image decomposition. According to the information characteristics of different components, use the improved wavelet critical domain method of this application to filter the high-frequency part of the image after wavelet decomposition, while the low-frequency part remains unchanged, so that the screen information in the scanning image is removed while retaining the detail part of the image. Then, perform wavelet reconstruction on the de-screened high-frequency part and low-frequency part. In the wavelet reconstruction process, use the weak wave model of wavelet decomposition to ensure that the deviation in the wavelet reconstruction process is as small as possible, obtain the gray image of each channel after de-screening, and then merge the three channels to obtain the de-screened color image. Finally, perform image super-resolution reconstruction on the de-screened color image to enhance the details and edges of the image;

[0013] Based on the technical solution of the improved method for wavelet critical domain de-screening in the above process, this application includes establishing wavelet decomposition levels, constructing weak wave models, improving the de-screening method of wavelet critical domains, estimating the variance of screen noise in wavelet critical domain filtering and calculating the size of the critical domain, and evaluating the de-screening effectiveness.

[0014] Preferably, it includes:

[0015] 1) Analyze the reasons for the formation of the screen in the scanning image: the digital gray-level technology used in the printing process and the setting of the scanning resolution size during the scanning process;

[0016] 2) Optimal combination of wavelet decomposition levels and weak wave models in the improved algorithm: Analyze the characteristics that the weak wave model needs to possess in wavelet critical domain de-screening, select the sym4 weak wave model, db4 weak wave model, coif2 weak wave model, and bior2.6 weak wave model respectively for de-screening processing, and use the first-level, second-level, and third-level wavelet decompositions for de-screening. Finally, determine that the first-level wavelet decomposition and db4 weak wave model achieve the best de-screening effect;

[0017] 3) Estimation of moiré noise in scanned images and improvement of wavelet critical regions: To address the deficiency of the method that only estimates the moiré noise in the first layer and determines the critical region in moiré removal based on wavelet critical regions, improvements are made in two aspects: estimating the moiré noise σ of each layer after the differentiation of the scanned image and determining the size of the wavelet critical region T for each layer.

[0018] 4) Super-resolution reconstruction of the moiré-removed image: Since the loss of image detail information and edge contour information occurs when performing moiré removal in the wavelet critical region on the scanned image, super-resolution reconstruction processing is carried out on the moiré-removed scanned image.

[0019] Preferably, improve the moiré removal method of the wavelet critical region: Correspondingly improve the deficiency in the estimation of the moiré noise variance σ. The expression of the modified noise variance σ is:

[0020]

[0021] In Equation 3, i refers to the wavelet differentiation level, and w HHi refers to the wavelet coefficient of the high-frequency detail signal in the i-th layer after wavelet differentiation;

[0022] The expression of the modified wavelet critical region T is:

[0023]

[0024] In Equation 4, N i represents the length of the wavelet in the i-th layer after wavelet differentiation;

[0025] Based on the expressions of the noise variance σ i and the wavelet critical region T i the estimated value of the noise variance σ of the improved scanned image i and the selection of the critical region T i are determined by the amount of moiré noise in each layer after wavelet differentiation. Among them, the selection of the critical region size has nothing to do with the selection of the weak wave model, and the selection of these two parameters affects the moiré removal effect of the scanned image respectively.

[0026] Preferably, establish the wavelet differentiation level: Based on the data evaluating the moiré removal effectiveness for different differentiation levels, when the wavelet differentiation level is higher, the data volume and clarity of the moiré-removed image are lower. According to the effect diagram of the moiré-removed image and the data evaluating the moiré removal effectiveness, the wavelet differentiation level is set to one level, two levels, and three levels.

[0027] Preferably, construct a weak wave model: Before adopting the weak wave model, confirm that it conforms to the following characteristics:

[0028] (1) Orthogonality: Let ψ(t) ∈ L 2 (R). If the function {ψ(t - k)} k∈Z meets:

[0029]

[0030] ψ(t) is a discrete wavelet function, belonging to the space L 2 (R), where t, l, and k are parameters, and the function {ψ(t - k)} k∈Z is an orthonormal system;

[0031] (2) Compact support: If the variable corresponding to f(t) is fixed as a constant 0 outside [a, b], then the function is said to have compact support on [a, b], and [a, b] is the support set of f(t);

[0032] (3) Regularity: It represents the smoothness degree of a function. The regularity of a function is related to its differentiability. If the regularity of a function is better, the higher the order of its differentiability, and the better the smoothness of the corresponding function curve;

[0033] (4) Vanishing moments: If the wavelet function ψ(t) ∈ L 2 (R) satisfies:

[0034]

[0035] Then the wavelet function ψ(t) has n - order vanishing moments. The smoother the image processed by the wavelet function y(t) using wavelet transform, the larger its vanishing moments, the smaller the high - frequency coefficients at high - resolution scales, and the more concentrated the energy of the corresponding wavelet - decomposed image;

[0036] (5) Symmetry: If the wavelet function used in the transformation is symmetric or anti - symmetric, the weak - wave model is called a symmetric weak - wave model;

[0037] The scanned image contains less moiré information. When performing wavelet transform, an orthogonal weak - wave model is used. When there is more moiré information, a biorthogonal weak - wave model is used for wavelet transform;

[0038] The sym4 weak - wave model, db4 weak - wave model, coif2 weak - wave model, and bior2.6 weak - wave model are respectively used to perform wavelet decomposition on the image with moiré information, and the effects of the image after moiré removal processed by each weak - wave model are compared.

[0039] Preferably, for the scanned data level: In an image, there are q kinds of gray - scale values of different pixel points, and the density of each gray - scale value distribution is respectively represented by P 1 , P 2 , …, P q to represent, and the data volume is expressed as:

[0040]

[0041] In Equation 5, H is the data volume of the image. When a = 2, the calculated data volume is in bits. If the probability of the gray value of each pixel point in the digital image is the same, the obtained H value is the largest. The scanned data level H characterizes the uncertainty of the image. For an image with the same gray value, its uncertainty is lower than that of a digital image with different gray values.

[0042] In the process of solving the color scanned data level in this application, the color image after descreening is first converted into a gray image, and then Equation 5 is used to calculate the data volume to obtain the data volume of the color digital image, which is specifically expressed as:

[0043]

[0044] The larger the value of the data volume, the greater the amount of information contained in the image, and the richer the detailed information of the image.

[0045] Preferably, texture recognition: The texture recognition is calculated based on the gradient harmonic operator. For a two-dimensional image function f(x, y), its corresponding transformation is defined as:

[0046]

[0047] The transformation calculation is linear. The two-dimensional gradient numerically realizes the addition of the second-order differential components of the two-dimensional function f(x, y) in the x direction, y direction, and diagonal direction. Its expression is:

[0048]

[0049] The calculation of Equation 8 is represented by a mask. When calculating the digital texture recognition of a size of M*N, it uses the gradient harmonic operator in the 3*3 neighborhood of a pixel to calculate the eight-neighborhood differential value of the pixel, and then adds the eight-neighborhood differential values of each pixel. Finally, the sum is divided by the size of the digital image to calculate the value of the texture recognition.

[0050] When calculating the color texture recognition in this application, the color image after descreening is first converted into a gray image, and then Equation 9 is used to solve the clarity of the image, and thus the clarity of the color digital image is obtained. Its expression is:

[0051]

[0052] If the digital image is more blurred, the change in gray value near the corresponding dot is also smaller, and the L value is smaller; on the contrary, if the clarity of the image is higher, it indicates that the contour of the digital image is more distinct, the change in gray value near the corresponding pixel is larger, and the calculated L value is larger, and the digital image is clearer.

[0053] Preferably, the evaluation scheme: there are two sizes of original manuscript samples of scanned images used in the evaluation, wherein the sizes of scanned images No. 1, 2, 3, 5 are 13.5cm×13.5cm, and the sizes of No. 4 and 6 are 6.5cm×6.5cm, the screen number of the printed image in the evaluation is 175lpi, and the scanning resolution is set to 300dpi;

[0054] Scanner descreening uses the NewColor7000 software that comes with the Heidelberg drum scanner D7100 to scan the color grayscale image, and uses the filter method in the NewColor7000 software to descreen the original image, without using the descreening function of NewColor7000 to obtain a scanned image containing a mesh pattern for descreening using the improved algorithm of this application, the wavelet critical method, and the Gaussian filter method of Photoshop;

[0055] When using Photoshop to achieve descreening, the Gaussian blur algorithm is used to descreen the color grayscale scanned image. The blur is to take the average value of the surrounding pixels for each pixel. If each middle point takes the average value of the 8 surrounding points, the point size becomes 1, which is equivalent to a blur effect. The middle point loses details. When calculating the middle point, it is necessary to calculate the weight value of each surrounding point. The calculation method is:

[0056]

[0057] Where σ represents the standard deviation. When the standard deviation σ = 0.6, the corresponding weight matrix and frequency response function;

[0058] The Gaussian blur algorithm in Photoshop is used to calculate the pixel value of the corresponding point in the image. When Photoshop is used for descreening, the radius of the Gaussian filter is set to 0.5 pixels. When USM sharpening is performed, the amount is set to 55%, the radius is set to 0.8 pixels, and the critical domain is set to 85 levels. Under these parameters, the descreening effect of the image scanned by this application using Photoshop is optimal.

[0059] Preferably, the multi-agent image super-resolution reconstruction algorithm is used to reconstruct the de-screened image. During the learning and training process, the images in the training set are selected from the database of the Image Research and Evaluation Laboratory of the University of California, Berkeley, USA. A total of 40 high-resolution images are selected for preprocessing in the training. During the preprocessing, the obtained intermediate-frequency information image and high-frequency information image are matched to form color block pairs, which are stored in the corresponding database for use in the reconstruction stage. Finally, the low-resolution image to be reconstructed in this application is input, and it is block-processed. The position of its nearest neighbor block in the corresponding model is found in the training database, and the missing high-frequency details are supplemented to the input low-resolution image to complete the image super-resolution reconstruction process.

[0060] Compared with the prior art, the innovation points and advantages of this application are as follows:

[0061] (1) This application has established an improved de-screening super-resolution reconstruction method for the gray level of color scanned images. First, the reasons for the formation of moiré patterns in scanned images are analyzed. The main reasons for the formation of moiré patterns in scanned images are the digital gray level technology used in the printing process and the setting of the scanning resolution during the scanning process. Second, the optimal combination of the wavelet differentiation level and the weak wave model in the improved algorithm is studied. By analyzing several characteristics that the weak wave model needs to possess in the wavelet critical domain de-screening, the sym4 weak wave model, db4 weak wave model, coif2 weak wave model, and bior2.6 weak wave model are selected for de-screening respectively, and the first-level, second-level, and third-level wavelet differentiations are used for de-screening. Through the analysis of evaluation data, it is finally found that the first-level wavelet differentiation and the db4 weak wave model can achieve the optimal de-screening effect. Third, the estimation of moiré noise in scanned images and the improvement of the wavelet critical domain are carried out. The deficiencies of the existing method for estimating and determining the critical domain only for the first-layer moiré noise in the wavelet critical domain de-screening are analyzed. Based on the existing wavelet critical domain de-screening method, this application improves in two aspects: estimating the moiré noise of each layer after the differentiation of the scanned image and determining the size of the wavelet critical domain of each layer, and good results are obtained. Fourth, the super-resolution reconstruction of the de-screened image is carried out. It is found through the change of the scanning data level and the clarity value during the evaluation process that when the scanned image is de-screened in the wavelet critical domain, not only the moiré information in the image is removed to a certain extent, but also the loss of image detail information and edge contour information is caused, resulting in the image becoming blurred to a certain extent. The super-resolution reconstruction process of the de-screened scanned image has achieved good results. The evaluation data shows that the improved method of this application can not only remove the moiré patterns in the scanned image to a certain extent, but also ensure the data volume and clarity of the de-screened image.

[0062] (2) This application uses different wavelet decomposition levels and different weak wave models to perform descreening on scanned images. When using the first-level wavelet decomposition and the db4 weak wave model for processing, the descreening effect is optimal. Then, the estimation of the screen noise σ and the selection of the wavelet critical domain T are improved, and the first-level wavelet decomposition and the db4 weak wave model are introduced into the improved algorithm. Finally, considering that using the improved algorithm of this application for descreening scanned images will further cause the loss of detail information and the blurring of the images to a certain extent, the super-resolution reconstruction of the descreened images is performed, and the descreened images are further enhanced. The method for improving the wavelet critical domain in descreening includes establishing the wavelet decomposition level and the weak wave model, estimating the variance of the screen noise in the wavelet critical domain filtering, calculating the size of the critical domain, and evaluating the descreening effect, realizing the accurate and efficient removal of screen from the scanned digital images, and ensuring that the digital images obtained during the digital processing of printed images can accurately and quickly transmit information.

[0063] (3) This application analyzes the reasons for the formation of screen in scanned images and designs the descreening of color grayscale scanned images, including the determination of the wavelet decomposition level and the weak wave model, the method for improving the wavelet critical domain, and the descreening processing of the wavelet critical domain. By analyzing the method for selecting the critical domain in the current existing wavelet critical domain descreening algorithms and improving their deficiencies, including improving the estimation of the screen noise in the scanned images and the selection of the wavelet critical domain. The scanned images are decomposed at different scales to obtain image information of different components, and then the improved wavelet critical domain is used for filtering processing, so that the descreened scanned images can be obtained. The super-resolution reconstruction of the descreened images is established to further enhance the detail information of the scanned images. For the evaluation of the descreening effect of color grayscale scanned images, two aspects of non-reference texture recognition and scan data level are selected to evaluate the descreening effect of the improved algorithm of this application, and it is compared with the descreening effects of the existing wavelet critical domain method, the filter method of the scanner's built-in NewColor7000 software, and the Gaussian blur algorithm of Photoshop, so as to reflect that the descreening effect of the algorithm of this application is good, and both the data volume and clarity have been greatly improved. Description of the Drawings

[0064] Figure 1 It is a flowchart of the method for improving descreening and super-resolution reconstruction of the grayscale of color scanned images.

[0065] Figure 2 It is a schematic diagram showing the influence of different decomposition levels and weak wave models on the descreening effect.

[0066] Figure 3 It is a comparison chart showing the influence of 4 different weak wave models on the descreening effect under the first-level wavelet decomposition.

[0067] Figure 4 It is a schematic diagram of the weight matrix and the frequency response curve.

[0068] Figure 5 It is a schematic diagram for comparing the effect of the improved algorithm and other methods in removing moiré patterns.

[0069] Figure 6 It is a schematic diagram showing the influence of different methods on the data volume and recognition rate of the moiré-removed image data.

[0070] Figure 7 It is a schematic diagram for comparing the moiré-removed image before and after super-resolution reconstruction. Specific implementation manners

[0071] The technical solution of the method for improving the gray level of color scanned images by moiré-removing super-resolution reconstruction provided by the present application will be further described below in conjunction with the accompanying drawings, so that those skilled in the art can better understand the present application and be able to implement it.

[0072] The present application performs moiré-removing processing on scanned images by using different wavelet decomposition levels and different weak wave models. The processing with the first-level wavelet decomposition and the db4 weak wave model has the best moiré-removing effect. Then, the estimation of the moiré noise σ and the selection of the wavelet critical domain T are improved, and the first-level wavelet decomposition and the db4 weak wave model are introduced into the improved algorithm. Finally, considering that using the improved algorithm of the present application to perform moiré-removing processing on scanned images will further cause loss of detail information and blurring of the images to a certain extent, the moiré-removed images are subjected to image super-resolution reconstruction, and further image enhancement processing is performed on the moiré-removed images.

[0073] The evaluation data of the present application show that the improved method of the present application can not only remove moiré patterns in scanned images to a certain extent, but also ensure the data volume and clarity of the moiré-removed images. Based on wavelet critical domain moiré-removing, the overall technical solution is as Figure 1 shown.

[0074] First, separate the channels of the color grayscale scanned image, and then process each of the separated R, G, and B color channels. The specific method is as follows: According to the moire distribution data of each color channel grayscale image, perform wavelet differentiation on the grayscale image of each color channel based on the optimal combination of the wavelet differentiation level and the weak wave model, so that the moire after image differentiation is most likely to be distributed on the high-frequency components of the image. According to the information characteristics of different components, use the improved wavelet critical domain method of this application to filter the high-frequency part of the image after wavelet differentiation, while keeping the low-frequency part unchanged, so as to remove the moire information in the scanned image and retain the detailed part of the image. Then, perform wavelet reconstruction on the high-frequency part and the low-frequency part after moire removal. During the wavelet reconstruction process, use the weak wave model of wavelet differentiation to ensure that the deviation in the wavelet reconstruction process is as small as possible, and obtain the grayscale image of each channel after moire removal. Then, merge the three channels to obtain the color image after moire removal. Finally, perform image super-resolution reconstruction on the color image after moire removal to enhance the details and edges of the image.

[0075] Based on the technical solution of the improved method for the wavelet critical domain in the above process, this application includes establishing the wavelet differentiation level and the weak wave model, estimating the moire noise variance in wavelet critical domain filtering, calculating the size of the critical domain, and evaluating the moire removal effect.

[0076] I. Establishing the wavelet differentiation level

[0077] Based on multiple evaluations of this application, for the scanned image of ordinary printed originals, performing first-level or second-level wavelet differentiation has achieved good moire removal effects. When using a higher level for differentiation and then performing wavelet reconstruction of the corresponding level on the image after moire removal, most of the detailed information of the image is lost. The main reason for this phenomenon is that during the wavelet reconstruction process, each time wavelet reconstruction is performed, there will be a certain degree of deviation in the reconstructed image. The greater the number of reconstructions, the greater the deviation.

[0078] Based on the evaluation data of the moire removal effect for different differentiation levels, when the level of wavelet differentiation is higher, the data volume and clarity of the image after moire removal are lower. According to the effect diagram of the moire removal image and the evaluation data of the moire removal effect, set the level of wavelet differentiation to one level, two levels, and three levels.

[0079] II. Constructing the weak wave model

[0080] Before using the weak wave model, confirm that it meets the following characteristics:

[0081] (1) Orthogonality: Let ψ(t) ∈ L 2 (R). If the function {ψ(t - k)} k∈Z meets:

[0082]

[0083] ψ(t) is a discrete wavelet function belonging to the space L 2 (R), where t, l, and k are parameters, and the function {ψ(t - k)} k∈Z is an orthonormal system;

[0084] (2) Compact support: If f(t) is fixed as a constant 0 outside the corresponding variable interval [a, b], then the function is said to have compact support on [a, b], and [a, b] is the support set of f(t);

[0085] (3) Regularity: It represents the smoothness of the function. The regularity of the function is related to its differentiability. The better the regularity of the function, the higher the order of its differentiability, and the smoother the corresponding function curve;

[0086] (4) Vanishing moments: If the wavelet function ψ(t) ∈ L 2 (R) satisfies:

[0087]

[0088] Then the wavelet function ψ(t) has n - order vanishing moments. The smoother the image processed by the wavelet function y(t) using wavelet transform, the larger its vanishing moments, the smaller the high - frequency coefficients under high - resolution scales, and the more concentrated the energy of the corresponding wavelet - differentiated image.

[0089] (5) Symmetry: If the wavelet function used in the transformation is symmetric or anti - symmetric, the weak - wave model is called a symmetric weak - wave model;

[0090] The scanned image contains less moiré information. When using wavelet transform with an orthogonal weak - wave model, when there is more moiré information, a biorthogonal weak - wave model is used for wavelet transform.

[0091] The sym4 weak - wave model, db4 weak - wave model, coif2 weak - wave model, and bior2.6 weak - wave model are respectively used to perform wavelet differentiation on the image with moiré information, and the effects of the images after moiré removal processed by each weak - wave model are compared.

[0092] III. Wavelet Differentiation Order and Weak - Wave Evaluation

[0093] To increase the universality of the improved method of this application, 6 different types of and different scanned images are selected for evaluation. The selection of the optimal combination of wavelet differentiation order and weak - wave model is based on the selection of the critical domain of the existing wavelet critical - domain moiré - removal method. The original data obtained from the 6 digital images in the evaluation are as Figure 2 shown.

[0094] For the same digital image, when using the same weak wave model to descreen the scanned image, as the number of wavelet decomposition levels increases, the data volume and clarity of the descreened image gradually decrease, especially the texture recognition ability decreases more significantly.

[0095] This phenomenon is caused by the loss of detailed information of the digital image during the process of using the wavelet critical domain to descreen. In addition, deviation will also occur during image reconstruction using wavelets. The more times of reconstruction, the greater the deviation. At the same time, the conclusion of the above analysis is also consistent with the result of subjective human eye evaluation, that is, the descreening effect of performing first-level wavelet decomposition on the scanned image is the best. Therefore, for this scanned image, when using the same weak wave model for descreening, the first-level wavelet decomposition has the best effect.

[0096] When descreening color grayscale scanned images using the same wavelet decomposition level, the descreening effects of the 4 weak wave models are not very different. In order to select the weak wave model with the best descreening effect for this evaluated scanned image, the following processing is done on the evaluation original data of 6 scanned images only at the first-level wavelet decomposition level: Add the scanned data levels and clarity values obtained by descreening the 6 scanned images at the first-level wavelet decomposition and the same weak wave model respectively, and then take the average value as the descreening effect value of this weak wave model at the first-level wavelet decomposition. This way of processing the original data not only avoids the influence of the contingency during the descreening process of the scanned image on the evaluation result, but also improves the stability and universality of the evaluation algorithm. According to the Figure 2 processing of the evaluation data at the first-level wavelet decomposition level in Figure 3 as shown.

[0097] According to Figure 3 the evaluation processed data in

[0098] At the first-level wavelet decomposition, the data volume values of the descreened images of the 4 different weak wave models are not very different. Among them, the value of the db4 weak wave model is the largest, the sym4 weak wave model is the second, the coif2 weak wave model is the third, and the value of bior2.6 is the smallest. There are certain differences in the values of texture recognition ability, and their value size ranking is the same as that at the scanned data level. When using the db4 weak wave model at the first-level wavelet decomposition, the data volume and clarity of the image are better than those of the other 3 weak wave models for descreening color grayscale scanned images.

[0099] IV. Descreening method for improving the wavelet critical domain

[0100] If the selection of the weak wave model is unreasonable, it will lead to serious loss of image information and distortion of the reconstructed image. For wavelet critical domain descreening, the determination of the critical domain size is also a key factor in determining the descreening effect of the image. If the critical domain is set too small, more moiré information will be retained in the image information, resulting in an unsatisfactory descreening effect of the image; on the contrary, if the critical domain is set too large, too much detail information of the scanned image will be lost, resulting in the blurring of the texture with important features in the scanned image. After wavelet reconstruction, the scanned image will be seriously distorted, making the subsequent processing of the scanned image more complicated. Therefore, it is very important to determine the critical domain size during the wavelet critical domain descreening process. Next, a method for selecting the critical domain is established, and an improved algorithm for the critical domain of this application is established according to this method.

[0101] This application improves the deficiencies of the existing unified critical domain method. In the existing unified critical domain descreening method, when estimating the moiré noise variance σ of the wavelet coefficients after differentiating the scanned image, since σ is only estimated for the first-layer detail signal after wavelet differentiation, it is inconsistent with the actual situation of the moiré information in the next-level wavelet differentiation, resulting in a certain error in determining the wavelet critical domain. When the wavelet critical domain is selected too large, too much image information will be removed, resulting in the blurring of the scanned image to a certain extent or the loss of detail information; when the wavelet critical domain is selected too small, it will also cause the moiré information in the scanned image to be not removed cleanly. Therefore, the inappropriate selection of the wavelet critical domain caused by inaccurate estimation of the moiré noise variance σ will lead to poor descreening effect of the scanned image.

[0102] Corresponding improvements are made to address the deficiencies in estimating the moiré noise variance σ according to the above existing method. The modified expression for the noise variance σ is:

[0103]

[0104] In Equation 3, i refers to the wavelet differentiation level, and w HHi refers to the wavelet coefficient of the high-frequency detail signal at the i-th layer after wavelet differentiation;

[0105] The modified expression for the wavelet critical domain T is:

[0106]

[0107] In Equation 4, N i represents the length of the i-th layer wavelet after wavelet differentiation;

[0108] Based on the expressions of the noise variance σ i and the wavelet critical domain T i the estimated value of the noise variance σ i of the improved scanned image and the critical domain T iThe selection is determined by the amount of moiré noise in each layer after wavelet decomposition. Among them, the selection of the critical domain size has nothing to do with the selection of the weak wave model. The selection of these two parameters respectively affects the moiré removal effect of the scanned image. Compared with before the improvement of the formula, not only the size of the noise variance in the moiré image is accurately estimated, making the actual situation of the moiré match the selection of the critical domain, but also the loss of image detail information is avoided, so that the moiré information in the scanned image is well removed, increasing the data volume of the image and improving the clarity of the image.

[0109] V. Evaluation of Moiré Removal Efficiency for Color Grayscale Scanned Images

[0110] This application performs moiré removal on the scanned image of the printed original. However, in most cases, the corresponding digital image of this printed original no longer exists. Therefore, the full-reference evaluation method cannot be used in this application. This application is based on the scanned data level and texture recognition.

[0111] (I) Scanned Data Level

[0112] In an image, there are q kinds of gray values for different pixel points, and the density of each gray value distribution is represented by P 1 , P 2 , …, P q respectively, and the data volume is expressed as:

[0113]

[0114] In Equation 5, H is the data volume of the image. When a = 2, the calculated data volume is in bits. If the possibility of the gray value of each pixel point in the digital image is the same, the obtained H value is the largest. The scanned data level H characterizes the uncertainty of the image. For an image with the same gray value size, its uncertainty is lower than that of a digital image with different gray value sizes;

[0115] In the process of solving the color scanned data level in this application, first convert the color image after moiré removal into a grayscale image, and then use Equation 5 to calculate the data volume to obtain the data volume of the color digital image, which is specifically expressed as:

[0116]

[0117] The larger the value of the data volume, the greater the amount of information contained in the image, and the richer the detail information of the image.

[0118] (II) Texture Recognition

[0119] Based on the gradient harmonic operator to calculate the texture recognition. For a two-dimensional image function f(x, y), its corresponding transformation is defined as:

[0120]

[0121] The transformation calculation is linear, and the two-dimensional gradient numerically realizes the addition of the second-order differential components of the two-dimensional function f(x,y) in the x direction, y direction, and diagonal direction. Its expression is:

[0122]

[0123] The calculation definition mask of formula 8 is used to represent that when calculating the digital texture recognition of a size of M*N, it uses the gradient reconciliation operator in the 3*3 neighborhood of a pixel to calculate the eight-neighborhood differential value of the pixel, and then adds the eight-neighborhood differential values ​​of each pixel, and finally divides the sum by the size of the digital image to calculate the value of texture recognition;

[0124] When calculating the color texture recognition degree in this application, the descreened color image is first converted into a grayscale image, and then the clarity of the image is solved by 9, so the clarity of the color digital image is obtained, and its expression is:

[0125]

[0126] If the digital image is more blurred, the grayscale value change near the corresponding dots is smaller, and the L value is smaller; on the contrary, if the image clarity is higher, indicating that the outline of the digital image is clearer, the grayscale value change near the corresponding pixels is greater, the calculated L value is larger, and the digital image is clearer.

[0127] VI. Analysis of evaluation results

[0128] 1. Evaluation plan

[0129] There are two sizes of original samples of scanned images used in this application evaluation. The sizes of scanned images No. 1, 2, 3, and 5 are 13.5cm×13.5cm, and the sizes of No. 4 and 6 are 6.5cm×6.5cm. The screen number of the printed image in the evaluation is 175lpi. In order to make the scanned image have obvious moiré information, the scanning resolution is set to 300dpi.

[0130] Scanner descreening uses the NewColor7000 software that comes with the Heidelberg drum scanner D7100 to scan the color grayscale image, and uses the filter method in the NewColor7000 software to descreen the original image. Instead of using the descreening function of NewColor7000, a scanned image containing a mesh pattern is obtained for descreening using the improved algorithm of this application, the wavelet critical method, and the Gaussian filter method of Photoshop.

[0131] When using Photoshop to remove moiré, the Gaussian blur algorithm is adopted to process the color grayscale scanned image. Blurring takes the average value of surrounding pixels for each pixel. If the average value of 8 surrounding points is taken for each middle point and the size of this point becomes 1, it is equivalent to producing a blurring effect and the details of the middle point are lost. When calculating the middle point, the weight value of each surrounding point needs to be calculated. The calculation method is as follows:

[0132]

[0133] Among them, σ represents the standard deviation. When the standard deviation σ = 0.6, the corresponding weight matrix and frequency response function are as Figure 4 shown.

[0134] When using the Gaussian blur algorithm in Photoshop to calculate the pixel value size of a corresponding point in the image, and then using Photoshop to remove moiré, the radius of the Gaussian filter is set to 0.5 pixels, while when performing USM sharpening, the amount is set to 55%, the radius is set to 0.8 pixels, and the threshold is set to 85 levels. Under these parameters, using Photoshop to remove moiré from the scanned image of this application has the best effect.

[0135] When using a Heidelberg drum scanner to scan a color grayscale scanned image, the filter function in its built-in NewColor7000 software can also achieve moiré removal. Through the evaluation of moiré removal for the color grayscale scanned image, it can be found that when the set moiré removal line count is consistent with the screen ruling of the grayscale printed original, the moiré in the scanned image can be removed to the greatest extent and a better moiré removal effect can be obtained.

[0136] Since the screen ruling of the evaluation object of this application is 175 lpi, so when using the filter function in NewColor7000 to remove moiré from the printed original in the evaluation, the set moiré removal line count is also 175 dpi, and the obtained moiré removal effect picture is obvious.

[0137] (2) Moiré removal effect analysis

[0138] When using the wavelet threshold to remove moiré from a color grayscale scanned image, selecting the db4 wavelet model for the first-level wavelet decomposition has the best effect. Therefore, in the improved algorithm of this application, the db4 wavelet model and the first-level wavelet decomposition are adopted, and the moiré removal effect of the improved algorithm is compared with that of the wavelet threshold method, the filter method of the scanner's built-in NewColor7000 software, and the Gaussian blur algorithm of Photoshop. The evaluation data of its scanning data level and clarity are as Figure 5 shown.

[0139] The de-screened images obtained by using the improved algorithm of this application have higher clarity and larger data volume than those of the existing wavelet threshold method, scanner filter method, and Gaussian blur algorithm in Photoshop after de-screening, that is, the de-screening effect of the improved algorithm of this application is better. At the same time, the conclusion of the above analysis is also consistent with the subjective evaluation of human eyes. To more intuitively show the superiority of the improved algorithm of this application, the Figure 5 evaluation data in it are plotted into a line chart to further reflect the effect of the improved algorithm of this application, as shown in Figure 6 .

[0140] By observing the line chart of the de-screening effects of the 4 methods in Figure 6 , it can be found that the data volume of the de-screened images obtained by applying the improved algorithm of this application is not much different from that of the de-screening effect of the Gaussian blur method in Photoshop, but both are better than the de-screening effect of the scanner filter method. And the clarity of the de-screened images obtained by applying the improved algorithm of this application is generally greater than the texture recognition rate obtained by the existing wavelet threshold de-screening method, scanner filter method, and Gaussian blur method in Photoshop, and the stability of its de-screening effect is also better than the other 3 methods.

[0141] According to the above analysis of the evaluation data, it can be found that using the method of improving the wavelet critical domain of this application to perform de-screening processing on color grayscale scanned images has a good de-screening effect, and is better than the de-screening effects of the wavelet threshold method, scanner filter method, and Gaussian blur method in Photoshop. Therefore, the de-screening method for color grayscale scanned images proposed in this application has certain significance and application value.

[0142] (III) Comparative Analysis after Super-Resolution Reconstruction of De-Screened Images

[0143] On the other hand, after the color grayscale scanned images pass through the improved de-screening algorithm of this application, to a certain extent, the details and edge contour information of the images are lost. Therefore, in order to further enhance the de-screened images, a multi-agent image super-resolution reconstruction algorithm is used to reconstruct the de-screened images.

[0144] During the learning and training process, the images in the training set are selected from the database of the Image Research and Evaluation Laboratory at the University of California, Berkeley, USA. A total of 40 high-resolution images are selected for preprocessing of training. During the preprocessing process, the obtained intermediate-frequency information images are matched with the high-frequency information images to form color block pairs one by one, and stored in the corresponding database for use in the reconstruction stage; finally, the low-resolution image to be reconstructed in this application is input, and it is divided into blocks. The position of its nearest neighbor block in the corresponding model is searched in the training database, and the lost high-frequency details are supplemented to the input low-resolution image, and the super-resolution reconstruction process of the image can be completed.

[0145] Now, compare the scan data level and clarity of the de-screened image before and after super-resolution reconstruction. The specific data is as Figure 7 shown.

[0146] After performing super-resolution reconstruction on the de-screened image, extract partial image information at corresponding positions before and after reconstruction. By calculating Figure 7 the data volume and clarity of the images in it, it can be found that the data volume and clarity of the super-resolution image after de-screening are much larger than those before reconstruction, which is consistent with the subjective evaluation effect of the human eye on the before-and-after effect diagrams of super-resolution reconstruction. Therefore, using super-resolution reconstruction processing of images can compensate for the loss of image detail information and edge contours during digital image processing, which is of great significance for image enhancement.

Claims

1. A method for improving the grayscale level of a color scanned image by de-screening and super-resolution reconstruction, characterized in that: The scanned image is de-screened using different wavelet differentiation levels and different weak wave models. The first-level wavelet differentiation and db4 weak wave model are used for processing, and the de-screening effect is optimal. Then the estimation of the moire noise σ and the selection of the wavelet critical domain T are improved, and the first-level wavelet differentiation and db4 weak wave model are introduced into the improved algorithm. Finally, based on the improved algorithm of this application, the de-screening of the scanned image will further cause the loss of detail information and the blurring of the image. The de-screened image is reconstructed with image super-resolution, and the de-screened image is further enhanced. Descreening is performed based on wavelet critical domain. First, the color grayscale scanned image is channel-separated, and then each color channel of R, G, and B obtained by separation is processed. The specific method is as follows: according to the reticulated distribution data of the grayscale image of each color channel, the grayscale image of each color channel is wavelet differentiated based on the optimal combination of wavelet differentiation series and weak wave model, so that the reticulated pattern after image differentiation is distributed on the high-frequency component of the image as much as possible. According to the information characteristics of different components, the high-frequency part of the image after wavelet differentiation is filtered by the improved wavelet critical domain method of the present application, while the low-frequency part remains unchanged, so that the reticulated information in the scanned image is removed and the details of the image are retained. Then, the high-frequency part and the low-frequency part after descreening are reconstructed by wavelet. In the process of wavelet reconstruction, the weak wave model of wavelet differentiation is used to ensure that the wavelet reconstruction process has as small a deviation as possible, and the grayscale image of each channel after descreening is obtained. Then, the three channels are merged to obtain the color image after descreening. Finally, the color image after descreening is processed based on image super-resolution reconstruction to enhance the details and edges of the image. Based on the technical solution for the improved wavelet critical domain descreening method in the above process, this application includes establishing wavelet differentiation series, constructing a weak wave model, improving the wavelet critical domain descreening method, estimating the variance of the descreening noise in the wavelet critical domain filtering and calculating the critical domain size, and evaluating the descreening effectiveness.

2. The color scanned image grayscale improvement descreening super-resolution reconstruction method according to claim 1, characterized in that: include: 1) Analyze the causes of the screen texture in the scanned image: the digital grayscale technology used in the printing process and the setting of the scanning resolution size in the scanning process; 2) Improve the optimal combination of wavelet differentiation levels and weak wave models in the algorithm: Analyze the characteristics of weak wave models in wavelet critical domain de-meshing, select sym4 weak wave model, db4 weak wave model, coif2 weak wave model and bior2.6 weak wave model for de-meshing respectively, and use primary, secondary and tertiary wavelet differentiation for de-meshing, and finally determine that primary wavelet differentiation and db4 weak wave model achieve the best de-meshing effect; 3) Estimation of scanned image moiré noise and improvement of wavelet critical domain: Based on the deficiency of the method that only estimates the first layer of moiré noise and determines the critical domain in wavelet critical domain de-screening, the moiré noise σ of each layer after the scanned image differentiation is estimated and the size of the wavelet critical domain T of each layer is determined; 4) Super-resolution reconstruction of de-screened images: Based on the loss of image detail information and edge contour information when the scanned image is de-screened using wavelet critical domain, super-resolution reconstruction is performed on the de-screened scanned image.

3. The method for grayscale improvement, descreening and super-resolution reconstruction of color scanned images according to claim 1, characterized in that: Improved wavelet critical region de-screening method: Corresponding improvements are made to the inadequacy of estimating the variance σ of the mesh noise. The expression of the modified noise variance σ is: In Equation 3, i refers to the wavelet differentiation level, w HHi Refers to the wavelet coefficient of the high-frequency detail signal in the i-th layer after wavelet differentiation; The expression of the modified wavelet critical region T is: In formula 4, N i Represents the length of the i-th layer wavelet after wavelet differentiation; Based on the noise variance σ i and wavelet critical region T i The expression of the improved scanning image noise variance σ i Estimation and critical region T i The selection is determined by the amount of mesh noise in each layer after wavelet differentiation. The selection of the critical domain size is independent of the choice of the weak wave model. The selection of these two parameters affects the descreening effect of the scanned image.

4. The method for grayscale improvement, descreening and super-resolution reconstruction of color scanned images according to claim 1, characterized in that: Establish wavelet differentiation levels: Based on the de-screening effectiveness evaluation data of different differentiation levels, when the wavelet differentiation level is higher, the data volume and clarity of the de-screened image are lower. According to the de-screening image effect diagram and de-screening effectiveness evaluation data, the wavelet differentiation level is set to level one, level two, and level three.

5. The color scanned image grayscale improvement descreening super-resolution reconstruction method according to claim 1, characterized in that: Construct a weak wave model: Before using the weak wave model, confirm that the following characteristics are met: (1) Orthogonality: Let ψ(t)∈L 2 (R), if the function {ψ(tk)} k∈Z conform to: ψ(t) is a discrete wavelet function belonging to the space L 2 (R), t, l, k are parameters, function {ψ(tk)} k∈Z is a canonical orthogonal system; (2) Compact support: If the variable [a, b] corresponding to f(t) is fixed to a constant 0, then the function is said to have compact support on [a, b], and [a, b] is the support of f(t); (3) Regularity: It indicates the smoothness of a function. The regularity of a function is related to its differentiability. If the regularity of a function is better, the number of differentiable functions is higher, and the corresponding function curve is smoother. (4) Vanishing moment: If the wavelet function ψ(t)∈L 2 (R) Satisfy: Then the wavelet function ψ(t) has an n-order vanishing moment. The smoother the image processed by the wavelet function y(t) using wavelet transform, the larger its vanishing moment, the smaller the high-frequency coefficient at high differentiation scale, and the more concentrated the energy of the corresponding image after wavelet differentiation. (5) Symmetry: The wavelet function used in the transformation is symmetric or antisymmetric, and the weak wave model is called a symmetric weak wave model; When the scanned image contains less reticulation information, the orthogonal weak wave model is used for wavelet transform. When the reticulation information is more, the biorthogonal weak wave model is used for wavelet transform. The sym4 weak wave model, db4 weak wave model, coif2 weak wave model and bior2.6 weak wave model are used to perform wavelet differentiation on images with texture information, and the effect of each weak wave model on the de-screened image is compared.

6. The method for improving grayscale descreening and super-resolution reconstruction of color scanned images according to claim 1, characterized in that: Scanning data level: In an image, there are q grayscale values ​​of different pixels, and the density of each grayscale value distribution is represented by P1, P2, ..., P q To express, the amount of data is expressed as: In formula 5, H is the data volume of the image. When a=2, the calculated data volume is in bits. If the grayscale value of each pixel of the digital image has the same probability of expression, the obtained H value is the largest. The scan data level H represents the uncertainty of the image. The uncertainty of an image with the same grayscale value is lower than that of a digital image with different grayscale values. In the process of solving the color scanning data level in this application, the descreened color image is first converted into a grayscale image, and then the data volume is calculated using Formula 5 to obtain the data volume of the color digital image, which is specifically expressed as: The larger the data volume is, the more information the image contains, and the richer the image details are.

7. The method for improving grayscale descreening and super-resolution reconstruction of color scanned images according to claim 1, characterized in that: Texture recognition: Texture recognition is calculated based on the gradient harmonic operator. For a two-dimensional image function f(x,y), the corresponding transformation is defined as: The transformation calculation is linear, and the two-dimensional gradient numerically realizes the addition of the second-order differential components of the two-dimensional function f(x,y) in the x direction, y direction, and diagonal direction. Its expression is: The calculation definition mask of formula 8 is used to represent that when calculating the digital texture recognition of a size of M*N, it uses the gradient reconciliation operator in the 3*3 neighborhood of a pixel to calculate the eight-neighborhood differential value of the pixel, and then adds the eight-neighborhood differential values ​​of each pixel, and finally divides the sum by the size of the digital image to calculate the value of texture recognition; When calculating the color texture recognition degree in this application, the descreened color image is first converted into a grayscale image, and then the clarity of the image is solved by 9, so the clarity of the color digital image is obtained, and its expression is: If the digital image is more blurred, the grayscale value change near the corresponding dots is smaller, and the L value is smaller; on the contrary, if the image clarity is higher, indicating that the outline of the digital image is clearer, the grayscale value change near the corresponding pixels is greater, the calculated L value is larger, and the digital image is clearer.

8. The color scanned image grayscale improvement descreening super-resolution reconstruction method according to claim 1, characterized in that: Evaluation plan: There are two sizes of original samples of scanned images used in the evaluation. The sizes of scanned images No. 1, 2, 3, and 5 are 13.5cm×13.5cm, and the sizes of No. 4 and 6 are 6.5cm×6.5cm. The screen number of the printed image in the evaluation is 175lpi, and the scanning resolution is set to 300dpi. Scanner descreening uses the NewColor7000 software that comes with the Heidelberg drum scanner D7100 to scan the color grayscale image, and uses the filter method in the NewColor7000 software to descreen the original image, without using the descreening function of NewColor7000 to obtain a scanned image containing a mesh pattern for descreening using the improved algorithm of this application, the wavelet critical method, and the Gaussian filter method of Photoshop; When using Photoshop to achieve descreening, the Gaussian blur algorithm is used to descreen the color grayscale scanned image. The blur is to take the average value of the surrounding pixels for each pixel. If each middle point takes the average value of the 8 surrounding points, the point size becomes 1, which is equivalent to a blur effect. The middle point loses details. When calculating the middle point, it is necessary to calculate the weight value of each surrounding point. The calculation method is: Where σ represents the standard deviation. When the standard deviation σ = 0.6, the corresponding weight matrix and frequency response function; The Gaussian blur algorithm in Photoshop is used to calculate the pixel value of the corresponding point in the image. When Photoshop is used for descreening, the radius of the Gaussian filter is set to 0.5 pixels. When USM sharpening is performed, the amount is set to 55%, the radius is set to 0.8 pixels, and the critical domain is set to 85 levels. Under these parameters, the descreening effect of the image scanned by this application using Photoshop is optimal.

9. The method for improving grayscale descreening and super-resolution reconstruction of color scanned images according to claim 1, characterized in that: A multi-agent image super-resolution reconstruction algorithm is used to reconstruct the de-networked image. During the learning and training process, the images in the training set are selected from the database of the Image Research and Evaluation Room of the University of Berkeley, USA. A total of 40 high-resolution images are selected for training preprocessing. During the preprocessing process, the obtained intermediate-frequency information image is matched with the high-frequency information image to form color block pairs, and stored in the corresponding database for use in the reconstruction stage; finally, the low-resolution image to be reconstructed in this application is input and processed in blocks. The position of its nearest neighbor block in the corresponding model is found in the training database, and the lost high-frequency details are supplemented to the input low-resolution image to complete the image super-resolution reconstruction process.

Images