Printing pattern background removing method and device for DTG and storage medium

Abstract

The invention discloses a printing pattern background removing method and device for DTG and a storage medium, and relates to the technical field of digital printing, and the method comprises the steps: converting an image into an LAB color space, carrying out the histogram equalization preprocessing, enhancing the image contrast and detail features, and improving the image quality. Performing joint morphological operation on a transparency channel matrix and an original image by utilizing multi-scale open operation and matching closed operation, retaining image details and optimizing edge transition, constructing an image pyramid by adopting a multi-resolution fusion technology, realizing cross-level feature fusion, improving an edge smoothing effect, and introducing an integrity detection mechanism to realize the edge smoothing effect. In addition, image sub-blocks are processed in parallel through a multi-thread or GPU acceleration technology, the calculation efficiency is improved, the problem of low-contrast image background removal is effectively solved, the printing quality and the production efficiency of printed patterns are remarkably improved, and the method has good application prospects.

Description

Technical Field

[0001] The present invention relates to the field of digital printing technology, and more specifically, to a method, device, and storage medium for removing the background of a printed pattern for DTG. Background Art

[0002] In scenarios such as customized clothing and industrial large-scale clothing printing, the application of digital direct-to-garment (DTG) printing technology is becoming increasingly popular. Generally, to save ink, improve printing efficiency, and ensure the clarity and aesthetics of printed patterns, it is necessary to remove the solid-color background of the printed pattern and only retain the foreground pattern. Currently, the image threshold segmentation method is a commonly used background removal method. Its principle is to select one or more thresholds based on the gray-scale or color distribution of the image and divide the pixels into two categories: background and foreground. For high-contrast images, this method can quickly and effectively separate the background and foreground, and the technology is mature and easy to implement. However, when processing low-contrast images, the color transition between the foreground and the background is relatively smooth, the difference in pixel gray-scale values is small, and traditional threshold segmentation techniques are difficult to accurately identify background pixels. If forced to apply, when the edge of the foreground image and the background image have low contrast, direct processing may result in edge jagging or distortion, affecting details. In addition, traditional methods lack a protection mechanism for image details when processing complex patterns, and are prone to losing some subtle but important pattern information, affecting the final printing quality.

[0003] To solve the above problems, researchers have conducted extensive explorations. Some methods attempt to improve the threshold selection algorithm, such as using the Otsu method, iterative method, etc. to dynamically determine the threshold to meet the background removal requirements of different images. However, these improvements are still limited to pixel judgment at a single scale and cannot comprehensively consider the multi-scale features and local characteristics of the image, and the improvement effect on low-contrast regions is limited. There are also some methods that introduce edge detection technology and repair the edges after removing the background. However, such methods often require additional computing resources, and the repair effect depends on the accuracy of edge detection, making it difficult to ensure the stable improvement of the overall image quality.

[0004] Therefore, the existing technology has problems such as edge jagging, detail loss, and low computational efficiency when dealing with the background removal of low-contrast images. Summary of the Invention

[0005] In order to overcome the problems of edge jagging, detail loss, and low computational efficiency existing in the prior art when dealing with the background removal of low-contrast images, the present invention discloses a method, device, and storage medium for removing the background of a printed pattern for DTG, which can effectively solve the above technical problems.

[0006] To solve the above technical problems, the technical solution of the present invention is as follows:

[0007] A method for removing the printing pattern background for DTG, comprising the following steps:

[0008] Input the image to be processed and convert the image into the standard format of the preset color space;

[0009] Perform multi-scale background pixel recognition on the image, separate the color channels of the image, judge whether each pixel is a background color according to the dynamic threshold, and generate an alpha channel matrix;

[0010] Select the corresponding structural element based on the edge gradient feature of the image, and perform joint morphological operations on the alpha channel matrix and the original image;

[0011] Adopt multi-resolution fusion technology to smooth the edges of the processed image, and output a printing special image file with an alpha channel;

[0012] Perform integrity detection on the alpha channel of the output image. If residual background pixels are detected, return to the step of performing multi-scale background pixel recognition on the image to readjust the threshold and process;

[0013] Divide the image into multiple sub-blocks, and use multi-threading or GPU acceleration technology to synchronously execute the steps from multi-scale background pixel recognition to edge smoothing on each sub-block, and finally merge the processing results.

[0014] Preferably, the dynamic threshold judgment specifically includes:

[0015] Calculate the initial background threshold T0 according to the global color gamut distribution of the image;

[0016] Divide the image into N×N sub-regions, and detect the chromaticity variance of each sub-region;

[0017] When the chromaticity variance of the sub-region exceeds the preset fluctuation threshold, use the local adaptive algorithm to update the judgment threshold of this region to T1, where T1 = T0×(1 + α·σ), α is the adjustment coefficient, and σ is the chromaticity standard deviation of the region.

[0018] Preferably, the local adaptive algorithm further includes:

[0019] When it is detected that the threshold difference between adjacent sub-regions exceeds the preset tolerance, start the boundary transition processing module, and generate a gradient band with a width of K pixels between adjacent sub-regions, where the transparency of the gradient band smoothly transitions according to the β·arctan(d) function, d is the distance to the boundary, and β is the transition coefficient.

[0020] Preferably, the execution of morphological optimization processing includes:

[0021] Perform multi-scale opening operations on the alpha channel matrix, and the structural element size set is the multi-scale size;

[0022] At the same time, perform morphological closing operation on the RGB channels of the original image. The number of operations n is determined by the image resolution and satisfies n = [log2(W×H) / 1000], where W and H are the number of pixels of the width and height of the image.

[0023] Preferably, it further includes edge enhancement processing:

[0024] Extract the Sobel edge feature map of the image after morphological closing operation;

[0025] Convolve and fuse the edge feature map with the transparency channel matrix to generate a corrected transparency channel.

[0026] Preferably, the conversion of the image to the standard format of the preset color space includes:

[0027] Convert the original RGB image to the XYZ space;

[0028] Convert to the LAB space through a non - linear transformation matrix;

[0029] Perform histogram equalization processing on the L channel.

[0030] Preferably, the edge smoothing of the processed image using the multi - resolution fusion technology includes:

[0031] Construct an image pyramid, including the original resolution layer, 1 / 2 down - sampling layer, and 1 / 4 down - sampling layer;

[0032] Perform edge - preserving filtering at each level;

[0033] Use wavelet transform for cross - level feature fusion, and the fusion weight is determined by the signal - to - noise ratio of each level.

[0034] Preferably, it further includes a pre - processing step:

[0035] Detect the suspected pattern area in the image. When the area of the area is less than the preset threshold, start the micro - structure enhancement module;

[0036] The micro - structure enhancement module generates high - resolution texture features through a generative adversarial network and performs α - blending with the original image.

[0037] An electronic device, including a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the above - mentioned printed pattern background removal method.

[0038] A computer - readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the above - mentioned printed pattern background removal method.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows: Through multi-scale background pixel recognition, this method can comprehensively consider the multi-scale features and local characteristics of the image, avoiding the limitations of single-scale judgment. At the same time, the initial background threshold is calculated based on the global color gamut distribution of the image, and when the chromaticity variance in the sub-region exceeds the preset fluctuation threshold, a local adaptive algorithm is used to update the judgment threshold in this region. This judgment method makes the recognition of background pixels more accurate, effectively reducing misjudgment and missed judgment, and avoiding the problem of detail loss caused by incomplete or excessive background removal at the source. After background removal, in order to optimize the image edge and retain details, this method selects corresponding structural elements based on the edge gradient features of the image, and performs joint morphological operations on the transparency channel matrix and the original image. The combination of multi-scale opening operation and matching closing operation can not only remove noise and isolated points in the image, but also maintain the edge integrity and detail features of the image. In addition, by extracting the Sobel edge feature map of the image and performing convolution fusion with the transparency channel matrix, a corrected transparency channel is generated, realizing the enhancement of the edge, making the image edge clearer and smoother, and solving the problem of edge sawtooth. To improve the edge smoothing effect, this method adopts a multi-resolution fusion technology, constructs an image pyramid, performs edge-preserving filtering at different levels, and conducts cross-level feature fusion through wavelet transform, which can make full use of the image information at different resolutions to achieve more refined edge smoothing processing while retaining the important details of the image. In addition, the introduced integrity detection mechanism detects the transparency channel of the output image. If residual background pixels are found, it automatically returns to readjust the threshold and process, ensuring the thoroughness and accuracy of background removal and further improving the image quality. Aiming at the problem of low computational efficiency, this method divides the image into multiple sub-blocks, and uses multi-threading or GPU acceleration technology to synchronously execute the steps of multi-scale background pixel recognition to edge smoothing for each sub-block, and finally combines the processing results. Through parallel processing, the multi-core resources of the computer are fully utilized, shortening the processing time and improving the overall computational efficiency, meeting the requirements of high-efficiency background removal technology in actual production. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary, and for those of ordinary skill in the art, without creative efforts, other implementation drawings can be obtained based on the provided drawings.

[0041] Figure 1 It is a flowchart of a method for removing the printed pattern background for DTG;

[0042] Figure 2It is a flowchart of a method for removing the printed pattern background for DTG. Detailed implementation manners

[0043] The attached drawings are only for illustrative purposes and should not be construed as a limitation of this patent;

[0044] To better illustrate this embodiment, some components in the attached drawings will be omitted, enlarged or reduced, which do not represent the dimensions of the actual product;

[0045] For those skilled in the art, it is understandable that some well-known structures and their descriptions in the attached drawings may be omitted.

[0046] The technical solutions of the present invention will be further described below with reference to the attached drawings and embodiments.

[0047] Embodiment

[0048] A computer equipped with an Intel Core i7-10700K processor, an NVIDIA GeForce RTX 3080 GPU, and 32 GB of memory.

[0049] Windows 10 operating system, Python 3.8 programming environment, with image processing libraries such as OpenCV, NumPy, scikit-image, and the PyTorch framework for implementing the generative adversarial network (GAN) part.

[0050] A method for removing the printed pattern background for DTG, please refer to Figure 1-2 , including the following steps:

[0051] Input the image to be processed and convert the image into a standard format in a preset color space;

[0052] Perform multi-scale background pixel recognition on the image, separate the color channels of the image, judge whether each pixel is a background color according to the dynamic threshold, and generate an alpha channel matrix;

[0053] Select corresponding structuring elements based on the edge gradient features of the image, and perform joint morphological operations on the alpha channel matrix and the original image;

[0054] Use multi-resolution fusion technology to smooth the edges of the processed image and output a printing-specific image file with an alpha channel;

[0055] Perform integrity detection on the alpha channel of the output image. If residual background pixels are detected, return to the step of performing multi-scale background pixel recognition on the image to readjust the threshold and process;

[0056] Divide the image into multiple sub - blocks, and use multi - threading or GPU acceleration technology to synchronously execute the above steps of multi - scale background pixel recognition to edge smoothing for each sub - block, and finally merge the processing results.

[0057] The dynamic threshold judgment specifically includes:

[0058] Calculate the initial background threshold T0 according to the global color gamut distribution of the image;

[0059] Divide the image into N×N sub - regions, and detect the chromatic variance of each sub - region;

[0060] When the chromatic variance of the sub - region exceeds the preset fluctuation threshold, use the local adaptive algorithm to update the judgment threshold of this region to T1, where T1 = T0×(1 + α·σ), α is the adjustment coefficient, and σ is the chromatic standard deviation of the region.

[0061] The local adaptive algorithm also includes:

[0062] When it is detected that the threshold difference between adjacent sub - regions exceeds the preset tolerance, start the boundary transition processing module to generate a gradient band with a width of K pixels between adjacent sub - regions, where the transparency of the gradient band is smoothly transitioned according to the β·arctan(d) function, d is the distance to the boundary, and β is the transition coefficient.

[0063] The execution of morphological optimization processing includes:

[0064] Perform multi - scale opening operations on the transparency channel matrix, and the set of structural element sizes is multi - scale sizes, such as {3×3, 5×5, 7×7};

[0065] At the same time, perform matching closing operations on the RGB channels of the original image, and the number of operations n is determined by the image resolution, satisfying n = [log2(W×H) / 1000], where W and H are the number of pixels of the width and height of the image.

[0066] It also includes edge enhancement processing:

[0067] Extract the Sobel edge feature map of the image after the closing operation;

[0068] Convolve and fuse the edge feature map with the transparency channel matrix to generate a corrected transparency channel.

[0069] The conversion of the image to the standard format of the preset color space includes:

[0070] Convert the original RGB image to the XYZ space;

[0071] Convert to the LAB space through a non - linear transformation matrix;

[0072] Perform histogram equalization processing on the L channel.

[0073] The edge smoothing of the processed image using the multi - resolution fusion technology includes:

[0074] Construct an image pyramid, including the original resolution layer, 1 / 2 down - sampling layer, and 1 / 4 down - sampling layer;

[0075] Perform edge - preserving filtering at each level;

[0076] Perform cross - level feature fusion using wavelet transform, and the fusion weights are determined by the signal - to - noise ratio of each level.

[0077] It also includes a pre - processing step:

[0078] Detect the suspected pattern area in the image. When the area of the area is less than the preset threshold, start the micro - structure enhancement module;

[0079] The micro - structure enhancement module generates high - resolution texture features through a generative adversarial network and performs α - blending with the original image.

[0080] An electronic device includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the above - mentioned printed pattern background removal method.

[0081] A computer - readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the steps of the above - mentioned printed pattern background removal method.

[0082] In a specific implementation,

[0083] Use the OpenCV library to read the tif - format image file to be processed.

[0084] Convert the image from the original RGB format to the standard format of a preset color space. The specific steps are as follows:

[0085] First, convert the RGB image to the XYZ color space. The conversion formula is:

[0086] X = 0.4124R+0.3576G + 0.1805B

[0087] Y = 0.2126R+0.7152G + 0.0722B

[0088] Z = 0.0193R+0.1192G + 0.9505B

[0089] Then, the XYZ is converted to the LAB color space through a non-linear transformation matrix, where the L channel represents luminance, and the A and B channels represent the chromaticity components of green - red and blue - yellow respectively. Histogram equalization is performed on the L channel to enhance the contrast of the image, making the subsequent background removal operation more accurate.

[0090] Statistically analyze the global color gamut distribution of the image in the LAB color space, with a focus on analyzing the luminance histogram of the L channel.

[0091] Calculate the initial background threshold T giobal , which is automatically determined by the Otsu algorithm and can divide the image pixels into two major categories: background and foreground, maximizing the variance between the two categories.

[0092] Divide the image into a sub-region grid of N×N (in this embodiment, N is taken as 16).

[0093] For each sub-region, calculate its chromaticity variance, that is, the average variance of the pixel values of the A and B channels.

[0094] When the chromaticity variance of the sub-region exceeds the preset fluctuation threshold, such as 0.05, the local adaptive algorithm is used to update the judgment threshold T of this region local , and the formula is:

[0095] T local = T giobal ×(1 + α·δ)

[0096] where α is the adjustment coefficient (taken as 0.8 in this embodiment), and σ is the standard deviation of the chromaticity of this sub-region.

[0097] Traverse each pixel of the image, separate the LAB color channels, and judge whether each pixel is a background color according to the dynamic threshold T of the sub-region where it is located local

[0098] If the pixel belongs to the background color, set the corresponding position in the transparency channel matrix to 0 (completely transparent); otherwise, set a certain transparency value (between 0 and 255) according to the similarity between the pixel and the background color to achieve a smooth transition.

[0099] Merge the generated transparency channel with the original RGB image to form an RGBA - format image, completing the preliminary background removal.

[0100] Select the corresponding structural element based on the edge gradient characteristics of the image. For example, for regions with relatively smooth edges, a circular structural element is used; for regions with rich edge details, a rectangular or cross - shaped structural element is used.

[0101] ​Perform multi-scale opening operations on the transparency channel matrix. The set of structural element sizes is {3×3, 5×5, 7×7}. The opening operation first performs erosion to remove small background noise points and then performs dilation to restore the continuity of the foreground region. Structural elements of different scales are used to process detail features of different sizes.

[0102] At the same time, perform matching closing operations on the RGB channels of the original image. The number of operations n is determined by the image resolution and satisfies:

[0103]

[0104] where W and H are the number of pixels in the width and height of the image, and round represents rounding to the nearest integer.

[0105] The closing operation first performs dilation to enhance the connectivity of the foreground region and then performs erosion to remove possible holes inside the foreground region, making the foreground region more compact and complete.

[0106] After the closing operation, use the Sobel operator to extract the edge feature map of the image, calculate the gradients in the horizontal and vertical directions respectively, and then synthesize the edge intensity map.

[0107] Convolve and fuse the edge feature map with the transparency channel matrix to generate a corrected transparency channel. The convolution kernel uses a 3×3 Gaussian kernel to achieve smooth edge transitions while highlighting edge details.

[0108] Construct an image pyramid containing the original resolution layer, 1 / 2 downsampling layer, and 1 / 4 downsampling layer, and perform edge-preserving filtering, such as Guided Filter, on the images at each level to retain edge details while smoothing the image.

[0109] Perform wavelet transform on the images processed at each level, decompose them into low-frequency and high-frequency subbands, calculate the fusion weights according to the signal-to-noise ratio of each level. The levels with higher signal-to-noise ratio have larger weights in the fusion. Weightedly fuse the low-frequency and high-frequency subbands of different levels, and obtain the finally fused image through inverse wavelet transform to achieve edge smoothing at multiple resolutions and further improve the image quality.

[0110] Perform integrity detection on the transparency channel of the output image. Use the connected component analysis method to identify possible remaining background pixel regions. If remaining background pixels are detected, record their position and size information.

[0111] According to the distribution of residual background pixels, return the multi-scale background pixel recognition step, and fine-tune the dynamic threshold of the corresponding area. For example, for areas with more residual background, appropriately reduce its local threshold to more strictly identify background pixels; for foreground detail areas that may be misjudged as background, appropriately increase the threshold to avoid excessive removal, and re-execute the background removal and subsequent processing processes until the transparency channel integrity detection passes.

[0112] Divide the image into multiple sub-blocks of equal size. In this embodiment, the size of the sub-blocks is set to 256×256 pixels to balance processing efficiency and memory occupancy. Using multi-threading technology, allocate a thread to each sub-block, and simultaneously execute the steps from multi-scale background pixel recognition to edge smoothing. For computationally intensive operations, such as texture feature generation of generative adversarial network (GAN), wavelet transform, etc., use GPU acceleration technology to significantly improve the processing speed through the CUDA parallel computing platform.

[0113] Finally, merge the processing results of each sub-block to generate a complete printed special image file after background removal. Compare the processed image with the original image and the results of other background removal methods to observe the retention of foreground pattern details, edge smoothness, thoroughness of background removal, etc.

[0114] Calculate objective metrics such as peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) of the processed image, compare with the original image, quantitatively evaluate the degree of image quality retention, and statistically calculate the accuracy and recall rate of background removal. The accuracy reflects the proportion of correctly removed background pixels, and the recall rate reflects the proportion of correctly retained foreground pixels, comprehensively measuring the performance of the method.

[0115] The same or similar reference numerals correspond to the same or similar components;

[0116] The terms describing the positional relationship in the drawings are only for illustrative purposes and should not be construed as a limitation of this patent;

[0117] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation manners here. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A method for removing the printed pattern background for DTG, characterized in that, It includes the following steps: Input the image to be processed and convert the image into the standard format of a preset color space; Perform multi-scale background pixel recognition on the image, separate the color channels of the image, judge whether each pixel is a background color according to a dynamic threshold, and generate an alpha channel matrix; Select corresponding structuring elements based on the edge gradient features of the image, and perform joint morphological operations on the alpha channel matrix and the original image; Use multi-resolution fusion technology to smooth the edges of the processed image, and output a printing-specific image file with an alpha channel; Perform integrity detection on the alpha channel of the output image. If residual background pixels are detected, return to the step of performing multi-scale background pixel recognition on the image to readjust the threshold and process; Divide the image into multiple sub-blocks, and use multi-threading or GPU acceleration technology to synchronously execute the above steps from multi-scale background pixel recognition to edge smoothing for each sub-block, and finally merge the processing results.

2. The printing pattern background removal method according to claim 1, wherein The specific dynamic threshold judgment includes: Calculate the initial background threshold T0 according to the global color gamut distribution of the image; Divide the image into N×N sub-regions, and detect the chromaticity variance of each sub-region; When the chromaticity variance of the sub-region exceeds the preset fluctuation threshold, use a local adaptive algorithm to update the judgment threshold of this region to T1, where T1 = T0×(1 + α·σ), α is an adjustment coefficient, and σ is the chromaticity standard deviation of the region.

3. The printing pattern background removal method according to claim 2, wherein The local adaptive algorithm also includes: When the threshold difference between adjacent sub-regions is detected to exceed the preset tolerance, start the boundary transition processing module to generate a gradient band with a width of K pixels between adjacent sub-regions, where the transparency of the gradient band is smoothed according to the β·arctan(d) function, d is the distance to the boundary, and β is the transition coefficient.

4. The printing pattern background removal method according to claim 1, characterized in that The execution of morphological optimization processing includes: Perform multi-scale opening operations on the alpha channel matrix, and the structuring element size set is multi-scale sizes; At the same time, perform matching closing operations on the RGB channels of the original image. The number of operations n is determined by the image resolution and satisfies n = [log2(W×H) / 1000], where W and H are the number of pixels in the width and height of the image.

5. The printing pattern background removal method according to claim 4, wherein, It also includes edge enhancement processing: Extract the Sobel edge feature map of the image after the closing operation; Convolve and fuse the edge feature map with the alpha channel matrix to generate a corrected alpha channel.

6. The printing pattern background removal method according to claim 1, characterized in that, The conversion of the image into the standard format of a preset color space includes: Convert the original RGB image to the XYZ space; Convert to the LAB space through a non-linear transformation matrix; Perform histogram equalization processing on the L channel.

7. The printing pattern background removal method according to claim 1, wherein The use of multi-resolution fusion technology to smooth the edges of the processed image includes: Construct an image pyramid, including the original resolution layer, 1 / 2 downsampling layer, and 1 / 4 downsampling layer; Perform edge-preserving filtering at each level; Use wavelet transform for cross-level feature fusion, and the fusion weight is determined by the signal-to-noise ratio of each level.

8. The printing pattern background removal method according to claim 1, wherein It also includes a preprocessing step: Detect the suspected pattern area in the image. When the area of the area is less than the preset threshold, start the micro-structure enhancement module; The micro-structure enhancement module generates high-resolution texture features through a generative adversarial network and performs alpha blending with the original image.

9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the printed pattern background removal method described in claims 1-8.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, it implements the steps of the printed pattern background removal method described in claims 1-8.

Images