Image amplification method and system based on gradient adaptive interpolation and noise suppression

Abstract

The invention provides an image amplification method for gradient adaptive interpolation and noise suppression, and belongs to the field of image processing before ink-jet printing. Extracting a sawtooth texture position of the gradient mask image as a noise point through a directional non-extreme suppression operator; performing median filtering on the noise points to generate an intermediate stage gradient image; self-adaptive interpolation processing is carried out according to the edge texture position of the image in the middle gradient stage; the interpolation processing result is corrected through gradient weight and distance weighting, and interference fringes are suppressed; and sending the amplified image block after interference removal into a buffer area. According to the image amplification method based on gradient adaptive interpolation and noise suppression, the processing efficiency is remarkably improved through intelligent blocking and parallel optimization, and sawtooth noise caused by high-power amplification is accurately eliminated by combining a hierarchical amplification strategy and a directional non-extreme suppression operator. And the adaptive interpolation algorithm based on gradient guidance effectively inhibits interference fringes while keeping edge sharpness, and realizes synchronous optimization of operation speed and visual quality.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of inkjet printing pre-press image processing, and particularly relates to an image magnification method and system based on gradient adaptive interpolation and noise suppression. BACKGROUND

[0002] Inkjet printing is a process of spraying ink droplets onto the surface of a substrate (paper, textiles, plastics, etc.) according to the dot matrix data provided by a computer. This technology has not only been increasingly demanded in industrial production such as commercial advertising and digital proofing, but also widely applied in professional fields such as engineering drawing and map printing, due to its fast printing speed and high image precision.

[0003] RIP (Raster Image Processor) software is the core of pre-press data processing. RIP image processing specifically refers to the process of interpreting input files as raster bitmaps, through color management, interpolation magnification, halftone processing, and finally outputting binary dot matrix files, mainly serving large-format high-precision inkjet printing operations, with much higher requirements for processing capacity and quality than desktop inkjet printing devices. Many companies have accumulated technology and innovated algorithms in this field, such as international well-known RIP software such as ONYX RIP, Caldera RIP, and domestic Maintop company. The algorithms of each link in the RIP process will have a great impact on the final rendering imaging. Among them, the interpolation magnification calculation, which involves a large number of floating-point multiplication operations for high-resolution reconstruction of images, often becomes the most time-consuming link and is the bottleneck of algorithm optimization and application deployment.

[0004] The interpolation magnification link in RIP is mainly to meet the increasingly high printing precision and width of inkjet printing. At higher printing precision, we can more accurately control the position of ink dots, while also bringing a lot of computational pressure to pre-press data processing. On the one hand, various super-resolution algorithms and image enhancement algorithms can significantly optimize the display effect of the image, generating clear tile edges, rather than the smooth transition caused by Bilinear linear interpolation; on the other hand, the complexity of these algorithms is often very high, making it difficult to be practically applied in large image processing, such as deep learning-based super-resolution algorithms, edge-based interpolation algorithms, and NEDI algorithms based on local covariance. Currently, the algorithms commonly used in commercial RIPs include nearest neighbor interpolation, B-spline interpolation, various Sinc window function interpolation, and edge detection-based adaptive interpolation, etc. For example, the Chinese patent with application publication number CN118605821A adopts a similar block processing flow as the present application, but only uses coordinate rounding to calculate the sampling points, which will produce obvious sawtooth textures at a large magnification.

[0005] Although the traditional interpolation algorithm can meet the production demand of fast calculation, various harmful textures are generated in the amplification process, which brings serious trouble in various fine lines and text printing operations. For example, the nearest neighbor interpolation method (or called no interpolation, zero order hold) is generally considered to be able to preserve the complete original image information and clear edges in the case of integer multiple amplification, but obvious blocky (or jagged) textures can be observed in the case of high magnification; due to the characteristics of the kernel function, the Bicubic and Lanczos interpolation methods will generate ringing textures (or Gibbs effect) around the originally clear edges and lines. Based on the above status analysis, we improve the following goals and propose a reliable and practical image interpolation and amplification algorithm: Goal 1: Eliminate various harmful interference textures caused by traditional linear interpolation, including blocky textures, blurred textures and ringing textures; the sharpness of clear and smooth text, lines and various general edge textures, and the layer details of the printing effect are more prominent; Goal 2: Fast processing speed, and preserving the original image information to ensure the authenticity and reliability of the printing result, which meets the actual printing business demand. SUMMARY

[0006] The purpose of the present application is to overcome the above-mentioned shortcomings, and to provide an image amplification method and system based on gradient adaptive interpolation and noise suppression.

[0007] In order to achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows: In a first aspect, the present application provides an image amplification method based on gradient adaptive interpolation and noise suppression, comprising the following steps: Load the preprocessed CMYK original image into the memory, calculate the target magnification image parameters, determine the block height of the magnification image and the thread pool configuration according to the memory size and the CPU core number, and obtain the blocked image block; When the magnification ratio of the image block is ≤2, the image block is put into the buffer area; when the magnification ratio of the image block is >2, the image block is magnified to 2 times of the original image by nearest neighbor interpolation, a gradient mask image is generated, and the sawtooth texture position of the gradient mask image is extracted as a noise point by a directional non-extreme suppression operator; Median filtering is performed on the noise points to generate an intermediate stage gradient image; According to the edge texture position of the intermediate gradient stage image, adaptive interpolation processing is performed; The interpolation processing result is corrected by gradient weight and distance weighting to suppress interference stripes, and a magnified image block after removing interference is obtained; The magnified image block after removing interference is sent to the buffer area for subsequent processing.

[0008] The step of when the magnification of the tile is ≤2, putting the tile into the buffer; when the magnification of the tile is >2, magnifying the tile to twice the original image through nearest neighbor interpolation, generating a gradient mask image, and through a directional non-extreme value suppression operator, extracting the sawtooth texture position of the gradient mask image as a noise point, when the magnification of the tile is >2, the method of subsequent operations is as follows: The tile is magnified to twice the original image through nearest neighbor interpolation to generate an initial magnified image; The Sobel operator is used to calculate the horizontal and vertical gradients of the initial magnified image, which are added and normalized to obtain a gradient mask image; The gradient mask image is processed through a directional local non-extreme value suppression operator to exclude horizontal and vertical edge pixels to obtain a sawtooth texture position mask image with noise frequency characteristics.

[0009] The Sobel operator is used to calculate the horizontal and vertical gradients of the initial magnified image, which are added and normalized to obtain a gradient mask image, and the specific formula is as follows:

[0010] Wherein, represents the normalized gradient amplitude, represents the image horizontal gradient component in the horizontal direction, represents the image vertical gradient component in the vertical direction; is the minimum value of the global gradient amplitude GM; is the maximum value of the global gradient amplitude GM.

[0011] The gradient mask image is processed through a directional local non-extreme value suppression operator to exclude horizontal and vertical edge pixels to obtain a sawtooth texture position mask image with noise frequency characteristics, and the specific formula is as follows:

[0012] Wherein, represents the normalized gradient amplitude at coordinates channel ; represents the corrected gradient amplitude after directional non-extreme suppression; represents the diagonal neighborhood of pixel ; represents the horizontal and vertical neighborhood of pixel ; and is an adaptive threshold value for controlling the strictness of non-extreme value suppression; , is a positive suppression threshold value, Gradient comparison for diagonal neighborhood; Gradient comparison for horizontal and vertical neighborhood.

[0013] In the step of median filtering the noise points to generate the intermediate stage gradient image, the specific formula is as follows:

[0014]

[0015] Wherein, represents the image after median processing, represents the median operation, represents the gradient amplitude of the filtered image.

[0016] In the step of adaptive interpolation according to the edge texture position of the intermediate gradient stage image, the nearest neighbor interpolation is directly used for the smooth area, and the high-order linear Lanczos kernel function is used for horizontal and vertical interpolation in turn for the edge texture area.

[0017] The method of using high-order linear Lanczos kernel function to perform horizontal and vertical interpolation in turn is as follows: For each row, fix The specific formula for horizontal interpolation in the horizontal direction is as follows:

[0018] Wherein, is the intermediate result of horizontal interpolation, representing the weighted pixel value in the kth row in the vertical direction; represents the pixel value of the ith sampling point, represents the horizontal coordinate displacement between the ith sampling point and the target interpolation point, represents the weight coefficient of Lanczos kernel function in the horizontal direction; is the window size coefficient; The formula for vertical interpolation in the vertical direction is as follows:

[0019] Wherein, is the vertical interpolation result, is the vertical direction offset, represents the weight coefficient of Lanczos kernel function in the vertical direction.

[0020] In the step of correcting the interpolation processing result with gradient weight and distance weighting to suppress interference stripes and obtain the interference-removed enlarged tile, the formula for correcting the interpolation processing result with gradient weight and distance weighting is as follows:

[0021]

[0022]

[0023] wherein, is a gradient adaptive weight, the greater the gradient, the smaller the weight; is a gradient attenuation coefficient, used to control the sensitivity of the gradient to the weight; is an intermediate gradient amplitude, used to represent the edge intensity of the pixel point k; is a nonlinear index; is a spatial distance weight, used to reflect the position correlation between the domain pixel k and the target interpolation point, the closer the distance, the greater the weight; is a gradient weighted interpolation result; is the pixel value of the pixel point k after median filtering.

[0024] The gradient weighted interpolation result is used to adaptively correct the Lanczos interpolation result to obtain an interference-removed enlarged tile:

[0025] wherein, is a final interpolation result, is a hybrid weight coefficient, is a gradient threshold, used to distinguish between edge regions and smooth regions; is the nearest interpolation result of the smooth region.

[0026] In a second aspect, the present application provides an image enlargement system with gradient adaptive interpolation and noise suppression, comprising: An image preprocessing module is used to load the preprocessed CMYK original image into memory, calculate the target enlarged image parameters, determine the block height of the enlarged image and the thread pool configuration according to the memory size and the CPU core number, and obtain the blocked tiles. A gradient mask image generation module is used to determine the magnification of the tiles, when the magnification of the tiles is ≤2, the tiles are put into the buffer area; when the magnification of the tiles is >2, the tiles are enlarged to 2 times the original image through nearest interpolation, a gradient mask image is generated, and the sawtooth texture position of the gradient mask image is extracted as a noise point through a directional non-extreme suppression operator. A median filtering module is used to median filter the noise points to generate an intermediate stage gradient image. An adaptive interpolation processing module is used to perform adaptive interpolation processing on the edge texture position of the intermediate gradient stage image. The correction module is used to correct the interpolation results with gradient weights and distance weights, suppress interference stripes, and obtain magnified patches after removing interference. The transmission module is used to send the enlarged image tiles after interference removal into the buffer for subsequent processing.

[0027] Compared with the prior art, the present invention has the following beneficial effects: This invention provides an image magnification method with gradient adaptive interpolation and noise suppression, comprising the following steps: loading a preprocessed CMYK original image into memory, calculating the parameters of the target magnified image, determining the block height and thread pool configuration of the magnified image based on the memory size and the number of CPU cores, and obtaining the block-wise image patches; when the magnification ratio of the image patch is ≤2, placing the image patch into a buffer; when the magnification ratio of the image patch is >2, magnifying the image patch to twice the size of the original image through nearest neighbor interpolation, generating a gradient mask image, and extracting the jagged texture positions of the gradient mask image as noise points using a directional non-extreme suppression operator; performing median filtering on the noise points to generate an intermediate stage gradient image; performing adaptive interpolation processing on the edge texture positions of the intermediate gradient stage image; correcting the interpolation processing results with gradient weights and distance weights to suppress interference stripes, obtaining magnified image patches after removing interference; and sending the magnified image patches after removing interference into a buffer for subsequent processing. The gradient-adaptive interpolation and noise suppression image magnification method significantly improves processing efficiency through intelligent block division and parallel optimization. It combines a hierarchical magnification strategy and a directional non-extreme suppression operator to accurately eliminate jagged noise caused by high magnification. Furthermore, the gradient-guided adaptive interpolation algorithm effectively suppresses interference stripes while preserving edge sharpness, achieving simultaneous optimization of computational speed and visual quality.

[0028] Furthermore, by dynamically determining the image segmentation strategy and thread pool configuration based on memory capacity and the number of CPU cores, precise scheduling of computing resources is achieved, significantly improving the efficiency of large-scale image processing, especially suitable for the real-time processing needs of high-resolution CMYK images. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the nearest neighbor interpolation method; Figure 3 This is a flowchart of the image processing process; Figure 4 A schematic diagram of a noise mask extracted using the directional suppression operator; Figure 5 A schematic diagram illustrating the calculation using a sliding window for linear interpolation; Figure 6 This is a schematic diagram of a gradient-based interpolation method. Figure 7This is a comparison chart showing the interpolation results after using the method of this invention; Figure 8 This is a halftone rendering of the result of the method of the present invention. Detailed Implementation

[0030] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0031] Example 1 like Figure 1 As shown, an image magnification method with gradient adaptive interpolation and noise suppression includes the following steps: S1: Load the preprocessed CMYK original image into memory, calculate the parameters of the target magnified image, determine the block height and thread pool configuration of the magnified image based on the memory size and the number of CPU cores, and obtain the block-wise image blocks; S2: When the magnification of the patch is ≤2, the patch is placed in the buffer; when the magnification of the patch is >2, the patch is magnified to twice the size of the original image by nearest neighbor interpolation to generate a gradient mask image. The jagged texture position of the gradient mask image is extracted as noise point by the directional non-extreme suppression operator. S3: Perform median filtering on noise points to generate an intermediate stage gradient image; S4: Perform adaptive interpolation processing on the edge texture position of the intermediate gradient stage image; S5: Correct the interpolation results with gradient weights and distance weights to suppress interference stripes and obtain magnified patches after removing interference; S6: Send the enlarged tile after removing interference into the buffer for further processing.

[0032] Specifically, in S1, the rasterized, color-separated, and color-linearized CMYK original image is loaded into memory. Based on the RIP parameters, the magnification factor and horizontal and vertical magnification factor of the target image are calculated according to the target resolution and image size. The appropriate image data block height is determined according to the computer memory size. The number of interpolation threads and the size of the halftone thread pool are determined according to the number of CPU cores.

[0033] The RIP parameters specifically refer to the following seven parameters:

[0034] in, The target interpolated pixel width and height are calculated from the horizontal and vertical resolutions (dpi) and the length (inch); Indicates the horizontal and vertical magnification factors; These refer to the vertical height of the tiles processed in batches and the number of interpolation and halftone threads, respectively.

[0035] Specifically, in S2, the image magnification is determined, and when the image magnification is greater than 2 times, the following steps are implemented: The tile is enlarged to twice the original image by nearest-neighbor interpolation to generate an initial enlarged image, and the specific formula is as follows:

[0036] wherein, represents the pixel value of the channel of the image after 2 times nearest-neighbor interpolation enlargement at the coordinate ; is the original input image.

[0037] The gradients in the horizontal and vertical directions of the initial enlarged image are calculated using the Sobel operator, which are added and normalized to obtain a 4-channel gradient mask image.

[0038] The gradient image is calculated using the operator based on the 2 times nearest-neighbor interpolation result. The convolution kernel of the operator in the direction is represented as:

[0039] The convolution kernel of the operator in the direction is represented as:

[0040] The absolute values of the gradients are added:

[0041] The gradients are normalized:

[0042] wherein, represents the normalized gradient amplitude, represents the gradient component of the image in the horizontal direction, represents the gradient component of the image in the vertical direction; is the minimum value of the global gradient amplitude GM; is the maximum value of the global gradient amplitude GM.

[0043] As preferred in this embodiment, the gradients are normalized to to better extract edges in different color channels, and the formula is as follows:

[0044] The gradient mask image is processed by a local non-maximum suppression operator with directionality to exclude horizontal and vertical edge pixels, to obtain a sawtooth texture position mask image with noise frequency characteristics, and the specific method is as follows: In the gradient image, the , edge and isolated pixel points are excluded to avoid being removed by mistake and ensure the integrity of the information; by adjusting the threshold of the horizontal and vertical non-maximum suppression operator, the position mask of the sawtooth noise is filtered, and the specific formula is as follows: ,

[0045] Among them, represents the normalized gradient amplitude at the coordinates , channel ; represents the corrected gradient amplitude after directional non-maximum suppression; represents the diagonal neighborhood of pixel ; represents the horizontal and vertical neighborhood of pixel ; and are adaptive thresholds for controlling the strictness of non-maximum suppression; , is a positive suppression threshold, which is selected as 25 and 10 in the test to obtain ideal results, is used for gradient comparison of the diagonal neighborhood; is used for gradient comparison of the horizontal and vertical neighborhood.

[0046] The filtered gradient image is shown in Figure 4 , and the corresponding image point is the nearest neighbor interpolation sawtooth texture generation point.

[0047] Specifically, in S3, the noise points of the position mask are extracted and median filtered to control the elimination of sawtooth edges, and an intermediate stage gradient image is generated, and the specific formula is as follows:

[0048]

[0049] Among them, represents the image after median processing, represents the median operation, represents the gradient amplitude of the filtered image.

[0050] Specifically, in S4, according to the intermediate stage gradient image, the edge texture position is judged and adaptive image processing is performed. The specific method is as follows: ​For smooth region, directly use nearest neighbor interpolation; For edge texture region, use high-order linear Lanczos function for interpolation, based on intermediate stage gradient input image, use Lanczos kernel function to perform horizontal and vertical interpolation in turn. Lanczos window length is set to 8, and at a large magnification, a sliding window can be used to realize fast interpolation operation.

[0051] The application is suitable for any multiple magnification requirements, so it is suitable to use linear interpolation as the basis of image magnification. For linear interpolation, generally, one-dimensional interpolation is calculated twice in horizontal and vertical directions, so that each row / column of the image to be interpolated can share common interpolation coefficients and sampling point index vectors, and the calculation is performed in advance:

[0052] In the case of Lanczos interpolation selecting interpolation kernel size , the coefficient vector and the calculation formula are:

[0053]

[0054]

[0055] wherein, , are the interpolation coefficient vectors with length k, is the offset of the point to be interpolated and the sampling point.

[0056] When calculating the point to be interpolated, the k coefficients and indexes corresponding to the point to be interpolated are found, and horizontal and vertical interpolation is performed, that is, the multiplication of the sampling point matrix B and the horizontal direction coefficient C is first performed, and then the multiplication of the vertical direction coefficient A is performed:

[0057]

[0058] Due to the large magnification, there are two cases for D corresponding to adjacent rows to be interpolated: (1) completely the same, in which case the horizontal direction interpolation does not need to be repeated; (2) the sampling window slides down one row, in which case only one row of D needs to be removed and updated. The sliding window of linear interpolation is shown in Figure 5 .

[0059] For each row, fix , and the specific formula for horizontal interpolation in the horizontal direction is as follows:

[0060]

[0061] wherein, is the intermediate result of horizontal interpolation, representing the weighted pixel value of the kth row in the vertical direction; represents the pixel value of the ith sampling point, represents the horizontal coordinate displacement between the ith sampling point and the target interpolation point, represents the weight coefficient of the Lanczos kernel function in the horizontal direction; is the window size coefficient, and here = 4.

[0062] The formula for vertical interpolation in the vertical direction is as follows:

[0063]

[0064] wherein, is the vertical interpolation result, is the vertical direction offset, represents the weight coefficient of the Lanczos kernel function in the vertical direction.

[0065] Specifically, in S5, the weight is calculated according to the gradient size and distance, the modified value is further calculated from the 2 times up-sampling intermediate result, the edge area Lanczos interpolation result is weighted and covered, and the interference stripes and blocky textures generated by linear interpolation are suppressed.

[0066]

[0067]

[0068]

[0069] wherein, is the gradient adaptive weight, and the greater the gradient, the smaller the weight; is the gradient attenuation coefficient, used to control the sensitivity of the gradient to the weight; is the intermediate gradient amplitude, used to represent the edge strength of the pixel point k; is a nonlinear index; is a spatial distance weight, used to reflect the position correlation between the field pixel k and the target interpolation point, and the closer the distance, the greater the weight; is the gradient weighted interpolation result; is the pixel value of the pixel point k after median filtering.

[0070] The gradient weighted interpolation result is used to adaptively correct the Lanczos interpolation result, and an enlarged tile with interference removed is obtained.

[0071]

[0072] wherein, is the final interpolation result, is the blending weight coefficient, is the gradient threshold value for distinguishing the edge region and the smooth region; is the nearest interpolation result of the smooth region.

[0073] By the joint modulation of the above gradient weight and the distance weight , the texture definition is preserved and the blocking effect is suppressed.

[0074] Specifically, in S6, the amplified tile after interference removal is sent to the buffer area for subsequent processing by the halftone thread pool.

[0075] Embodiment 2 An image magnification system with gradient adaptive interpolation and noise suppression, comprising: An image preprocessing module for loading the preprocessed CMYK original image into memory, calculating the target magnification image parameters, determining the block height of the magnification image and the thread pool configuration according to the memory size and the number of CPU cores, and obtaining the blocked tile; A gradient mask image generation module for judging the magnification ratio of the tile, when the magnification ratio of the tile is ≤2, the tile is put into the buffer area; when the magnification ratio of the tile is >2, the tile is magnified to 2 times the original image by nearest neighbor interpolation, a gradient mask image is generated, and the sawtooth texture position of the gradient mask image is extracted as a noise point by a directional non-extreme suppression operator; A median filter module for median filtering the noise point to generate an intermediate stage gradient image; An adaptive interpolation processing module for adaptive interpolation processing of the edge texture position of the intermediate gradient stage image; A correction module for correcting the interpolation processing result with gradient weight and distance weighting to suppress interference stripes and obtain the magnified tile after interference removal; A transmission module for sending the magnified tile after interference removal to the buffer area for subsequent processing.

[0076] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application and not to limit them, although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, any modification or equivalent replacement thereof should be covered within the protection scope of the claims of the present application.

Claims

1. An image magnification method using gradient adaptive interpolation and noise suppression, characterized in that, Includes the following steps: The preprocessed CMYK original image is loaded into memory, the parameters of the target magnified image are calculated, and the block height and thread pool configuration of the magnified image are determined according to the memory size and the number of CPU cores to obtain the block-wise image. When the magnification of the patch is ≤2, the patch is placed in the buffer; when the magnification of the patch is >2, the patch is magnified to twice the size of the original image by nearest neighbor interpolation to generate a gradient mask image. The jagged texture position of the gradient mask image is extracted as noise point by the directional non-extreme suppression operator. Median filtering is applied to noise points to generate intermediate stage gradient images; Adaptive interpolation is performed on the image based on the edge texture position of the intermediate gradient stage image. The interpolation results are corrected by gradient weights and distance weights to suppress interference stripes and obtain magnified patches after removing interference. The enlarged image tile, after removing interference, is sent to the buffer for further processing.

2. The image magnification method with gradient adaptive interpolation and noise suppression according to claim 1, characterized in that, In the step of placing the patch in a buffer when the patch magnification is ≤2 and magnification is >2, the patch is magnified to twice the size of the original image using nearest neighbor interpolation to generate a gradient mask image, and the jagged texture positions of the gradient mask image are extracted as noise points using the directional non-extreme suppression operator, the method for subsequent operations when the patch magnification is >2 is as follows: The tile is enlarged to twice the size of the original image using nearest-neighbor interpolation to generate an initial enlarged image; The Sobel operator is used to calculate the horizontal and vertical gradients of the initial magnified image, which are then summed and normalized to obtain the gradient mask image. By processing the gradient mask image with a directional local non-extreme suppression operator, and excluding horizontal and vertical edge pixels, a jagged texture location mask image with noise frequency characteristics is obtained.

3. The image magnification method with gradient adaptive interpolation and noise suppression according to claim 2, characterized in that, The Sobel operator is used to calculate the horizontal and vertical gradients of the initial magnified image, which are then summed and normalized to obtain the gradient mask image. The specific formula is as follows: in, This represents the normalized gradient magnitude. Representing an image The gradient component in the horizontal direction, Representing an image The gradient component in the vertical direction; This represents the minimum value of the global gradient magnitude GM. This represents the maximum value of the global gradient magnitude GM.

4. The image magnification method with gradient adaptive interpolation and noise suppression according to claim 3, characterized in that, The gradient mask image is processed by a directional local non-extremum suppression operator to exclude horizontal and vertical edge pixels, resulting in a jagged texture location mask image with noise frequency characteristics. The specific formula is as follows: in, Indicates coordinates ,aisle Normalized gradient magnitude at point; This represents the corrected gradient magnitude after directional nonpolar suppression. Represents pixels The diagonal neighborhood; Represents pixels Horizontal and vertical neighborhoods; and An adaptive threshold is used to control the strictness of non-extreme value suppression; , The positive suppression threshold is... Gradient comparison for diagonal neighborhood; Used for gradient comparison in horizontal and vertical neighborhoods.

5. The image magnification method with gradient adaptive interpolation and noise suppression according to claim 1, characterized in that, The specific formula for the step of performing median filtering on noise points to generate the intermediate stage gradient image is as follows: in, This represents the image after median processing. This represents median operations. This represents the gradient magnitude of the filtered image.

6. The image magnification method with gradient adaptive interpolation and noise suppression according to claim 1, characterized in that, In the step of adaptively interpolating the image based on the edge texture position of the intermediate gradient stage image, nearest neighbor interpolation is used directly for smooth regions, while higher-order linear Lanczos kernel functions are used to perform horizontal and vertical interpolation sequentially for edge texture regions.

7. The image magnification method with gradient adaptive interpolation and noise suppression according to claim 1, characterized in that, The method of performing horizontal and vertical interpolation sequentially using a higher-order linear Lanczos kernel function is as follows: For each row, fixed The specific formula for horizontal interpolation in the horizontal direction is as follows: in, This is an intermediate result of horizontal interpolation, representing the weighted pixel value in the k-th row of the vertical direction; This represents the pixel value of the i-th sampling point. This represents the horizontal coordinate displacement between the i-th sampling point and the target interpolation point. This represents the weighting coefficients of the Lanczos kernel function in the horizontal direction; This is a window size factor; The formula for vertical interpolation is as follows: in, This is the result of vertical interpolation. This represents the vertical offset. This represents the weighting coefficient of the Lanczos kernel function in the vertical direction.

8. The image magnification method with gradient adaptive interpolation and noise suppression according to claim 1, characterized in that, In the step of correcting the interpolation result using gradient weights and distance weights to suppress interference fringes and obtain the enlarged image patch after removing interference, the formula for correcting the interpolation result using gradient weights and distance weights is expressed as follows: in, The weights are gradient-adaptive; the larger the gradient, the smaller the weight. This is the gradient decay coefficient, used to control the sensitivity of the gradient to the weights; This is the intermediate gradient magnitude, used to characterize the edge strength of pixel k; It is a non-linear exponent; This is the spatial distance weight, used to reflect the positional correlation between the neighborhood pixel k and the target interpolation point; the closer the distance, the greater the weight. This is the result of gradient-weighted interpolation; The pixel value is the median value of pixel k after median filtering.

9. The image magnification method with gradient adaptive interpolation and noise suppression according to claim 8, characterized in that, Adaptively correct the Lanczos interpolation results using gradient-weighted interpolation to obtain enlarged patches after removing interference: in, For the final interpolation result, For mixed weighting coefficients, This is the gradient threshold used to distinguish between edge regions and smooth regions; This is the nearest neighbor interpolation result for the smooth region.

10. An image magnification system with gradient adaptive interpolation and noise suppression as described in any one of claims 1 to 9, characterized in that, include: The image preprocessing module is used to load the preprocessed CMYK raw image into memory, calculate the parameters of the target magnified image, determine the block height and thread pool configuration of the magnified image based on the memory size and the number of CPU cores, and obtain the block-wise image tiles. The gradient mask image generation module is used to determine the magnification of the patch. When the magnification of the patch is ≤2, the patch is placed in the buffer. When the magnification of the patch is >2, the patch is magnified to twice the size of the original image by nearest neighbor interpolation to generate a gradient mask image. The jagged texture positions of the gradient mask image are extracted as noise points by the directional non-extreme suppression operator. The median filtering module is used to perform median filtering on noise points and generate intermediate stage gradient images. The adaptive interpolation module is used to perform adaptive interpolation processing on the image based on the edge texture position of the intermediate gradient stage image. The correction module is used to correct the interpolation results with gradient weights and distance weights, suppress interference stripes, and obtain magnified patches after removing interference. The transmission module is used to send the enlarged image tiles after interference removal into the buffer for subsequent processing.

Images