Ink-jet printing image color uniformity compensation method and device based on deep learning

Abstract

The invention discloses an ink-jet printing image color uniformity compensation method and device based on deep learning. The method comprises the following steps: step 1, data acquisition and preprocessing; 2, constructing a reverse compensation network RCN, inputting the original CMYK image into the RCN, and outputting a compensated CMYK image; 3, constructing a printer distortion simulation network PDM, inputting the compensated CMYK image into the PDM, and outputting an RGB image for simulating nonlinear color distortion generated in the printing and scanning process; 4, macroscopic uniformity detection is carried out on the RGB image output in the step 3; and 5, constructing a total loss function and carrying out joint training. According to the invention, the pixel level, the structure level, the perception level, the macroscopic uniformity, the color feature and other levels of the ink-jet printing image can be optimized, and the quality of the printed product is improved.

Description

Technical Field

[0001] The present invention relates to image processing technology, and in particular to a method and device for compensating the color uniformity of inkjet-printed images based on deep learning. Background Art

[0002] In the existing inkjet printing technology, due to reasons such as equipment hardware and ink characteristics, color non-uniformity problems often occur in printed images. Especially when printing a uniform cyan image, periodic light and dark stripe phenomena will appear in the vertical direction. To improve this problem, color compensation of the input image is usually carried out manually or in a fixed mode in advance, but it is impossible to accurately capture the non-linear distortion law in the printer output, and it is difficult to meet the requirements of the macroscopic uniformity score evaluation based on the measurement of a scanning spectrophotometer (such as the ISO TS 18621-21 standard). Therefore, there is an urgent need for a deep learning method and device that is data-driven, automatically learns the distortion law of the printer output, and incorporates macroscopic uniformity optimization in end-to-end training. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and device for compensating the color uniformity of inkjet-printed images based on deep learning in view of the deficiencies of the prior art, which can automatically learn the non-linear distortion generated during the process of converting a CMYK image into an RGB image by a printer, and perform reverse compensation on the input CMYK image, so that the finally output image after passing through the printer can achieve the expected effect in macroscopic uniformity, thereby meeting the requirements of the ISO TS 18621-21 standard.

[0004] The purpose of the present invention is achieved by the following technical solutions:

[0005] According to the first aspect of this specification, a method for compensating the color uniformity of inkjet-printed images based on deep learning is provided, and the method includes the following steps:

[0006] (1) Data acquisition and preprocessing: Generate an original CMYK image, and each channel maintains a uniform color output; output the generated original CMYK image through an inkjet printer, and use a scanner to obtain the printed RGB image; perform a color space conversion from the scanned RGB image to the XYZ space, and calculate the optical density OD value of each color feature;

[0007] (2) Construct a reverse compensation network RCN, input the original CMYK image into the RCN, and output the compensated CMYK image;

[0008] (3) Construct a printer distortion simulation network PDM, input the compensated CMYK image into the PDM, and output an RGB image to simulate the non-linear color distortion generated during the printing and scanning processes;

[0009] (4) Perform macroscopic uniformity detection on the RGB image output in step (3);

[0010] (5) Construct a total loss function including pixel-level loss, structural similarity loss, perceptual loss, macroscopic uniformity loss, and OD loss of different color features, and conduct joint training to finally optimize the inkjet printed image at the pixel level, structural level, perceptual level, macroscopic uniformity, and color feature levels.

[0011] Further, step (1) is specifically: generate a batch of original CMYK images with a size of 1054×884 pixels, conforming to the A4 paper ratio, and the cyan, magenta, yellow, and black channels all maintain uniform color output; output the generated original CMYK images through an inkjet printer, and use a scanner to obtain the printed RGB image; perform color space conversion from the RGB image to the XYZ space, and calculate the optical density OD values of each color;

[0012] The formula for converting the color space of the RGB image to the XYZ space is as follows:

[0013] X = 0.4124R + 0.3576G + 0.1805B

[0014] Y = 0.2126R + 0.7152G + 0.0722B

[0015] Z = 0.0193R + 0.1192G + 0.9505B

[0016] The formula for calculating the OD value of each color is as follows:

[0017] Cyan feature:

[0018] Magenta feature:

[0019] Yellow feature:

[0020] Further, step (2) is specifically: construct a reverse compensation network RCN, which adopts a U-Net architecture, including an encoder, a bottleneck layer, and a decoder. The input is the original CMYK image, and the output is the compensated CMYK image; the encoder part contains multiple convolutional blocks, each convolutional block contains two layers of convolution and activation functions, followed by a pooling layer; the bottleneck layer contains multiple residual blocks; the decoder part gradually restores the spatial dimension through upsampling and skip connections, and finally outputs the compensated CMYK image through a convolutional layer.

[0021] Further, step (3) is specifically as follows: Construct a printer distortion simulation network PDM, which adopts a U-Net architecture and includes a network input end, an encoder, a bottleneck layer, and a decoder. The input is the compensated CMYK image, and the output is the RGB image after simulating the printing and scanning processes. A convolutional layer is set at the network input end to convert the CMYK image into an internal feature representation. The encoder part contains multiple convolutional blocks, each convolutional block contains two layers of convolution and activation functions, and is followed by a pooling layer. The bottleneck layer contains multiple residual blocks. The decoder part outputs the RGB image through upsampling and skip connections.

[0022] Further, step (4) is specifically as follows: Perform regional processing on the RGB image output by the PDM, calculate the color mean and standard deviation of each region, statistically calculate the macroscopic uniformity score of the entire image, and construct a macroscopic uniformity loss.

[0023] The macroscopic uniformity loss is defined as follows: Divide the image into K regions and calculate the average color value of each region. Let μ i be the average color value of the i-th region, and be the average of the means of all regions. Then the calculation formula for the macroscopic uniformity loss is:

[0024]

[0025] After the scanned image is converted by XYZ, calculate the OD values of cyan, magenta, and yellow respectively, denoted as to obtain the true OD C , OD M , OD Y graph; Perform the same processing on the RGB image output by the PDM, that is, calculate the OD values of cyan, magenta, and yellow respectively after XYZ conversion, denoted as to obtain the predicted OD C , OD M , OD Y graph; Construct three OD losses respectively:

[0026] Cyan loss:

[0027] Magenta loss:

[0028] Yellow loss:

[0029] Further, step (5) is specifically as follows: Construct a total loss function and use the following combined loss function for training:

[0030]

[0031] Among them, is the pixel-level loss; is the structural similarity loss; is the perceptual loss; is the macroscopic uniformity loss; are the OD losses of the cyan, magenta, and yellow features respectively; λ SSIM , λ perc , λ uin , are the weights of each loss term.

[0032] Furthermore, the calculation formula of the pixel-level loss is:

[0033]

[0034] Among them, I pred (i), I true (i) represent the pixel values of the predicted image and the ground truth image at pixel i respectively, and N is the total number of pixels in the image;

[0035] The calculation formula of the structural similarity loss is:

[0036]

[0037] Among them, the structural similarity SSIM index measures the similarity of images in terms of brightness, contrast, and structure, and its formula is:

[0038]

[0039] Among them, μ x , μ y represent the mean values of the intensity values of each pixel in the defined area of the predicted image and the ground truth image respectively, and are the local variances, σ xy is the covariance, and C1 and C2 are constants set to avoid the denominator being zero;

[0040] The calculation formula of the perceptual loss is:

[0041]

[0042] Among them, represent the feature maps extracted by the pre-trained networks of the predicted image and the ground truth image at the l-th layer respectively.

[0043] Further, in step (5), through end-to-end joint training, the entire network composed of RCN and PDM is regarded as a whole. The overall loss is calculated through one forward propagation, and then the gradients of both parts are calculated simultaneously through backpropagation. Then, the gradient descent method is used to synchronously update all parameters, ultimately optimizing the inkjet printed image at all levels such as pixel level, structure level, perception level, macroscopic uniformity, and color characteristics.

[0044] According to the second aspect of this specification, there is provided an inkjet printed image color uniformity compensation device based on deep learning, including a memory and one or more processors. An executable code is stored in the memory. When the processor executes the executable code, it is used to implement the inkjet printed image color uniformity compensation method based on deep learning as described in the first aspect.

[0045] According to the third aspect of this specification, there is provided a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, it is used to implement the inkjet printed image color uniformity compensation method based on deep learning as described in the first aspect.

[0046] The beneficial effects of the present invention are as follows: The present invention provides an inkjet printed image color uniformity compensation method and device based on deep learning. By constructing an end-to-end model of a reverse compensation network RCN and a printer distortion simulation network PDM, and combining a macroscopic uniformity evaluation module, it realizes the optimized compensation for the color uniformity of the printed output image (in accordance with the ISO TS18621-21 standard). Ultimately, the inkjet printed image is optimized at all levels such as pixel level, structure level, perception level, macroscopic uniformity, and color characteristics, improving the quality of printed products. Description of the Drawings

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1 is a flowchart of the inkjet printed image color uniformity compensation method based on deep learning of the present invention.

[0049] Figure 2 is a schematic structural diagram of the reverse compensation network RCN of the present invention.

[0050] Figure 3 is a schematic structural diagram of the printer distortion simulation network PDM of the present invention.

[0051] Figure 4It is a printed and scanned image of the original cyan image.

[0052] Figure 5 It is a printed and scanned image of the optimized cyan image.

[0053] Figure 6 It is a comparison chart of the macroscopic uniformity scores of a part of the cyan image before and after optimization.

[0054] Figure 7 It is a structural diagram of the device for compensating the color uniformity of inkjet-printed images based on deep learning according to the present invention. Detailed implementation manners

[0055] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe in detail the specific implementation manners of the present invention with reference to the accompanying drawings.

[0056] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0057] The present invention provides a method for compensating the color uniformity of inkjet-printed images based on deep learning, as Figure 1 shown, which includes the following five steps:

[0058] Step 1: Data acquisition and preprocessing;

[0059] Step 2: Construct a reverse compensation network RCN, input the original CMYK image into the RCN, and output the compensated CMYK image;

[0060] Step 3: Construct a printer distortion simulation network PDM, input the compensated CMYK image output in Step 2 into the PDM, and output an RGB image to simulate the non-linear color distortion generated during the printing and scanning processes;

[0061] Step 4: Use a macroscopic uniformity evaluation module to perform macroscopic uniformity detection on the RGB image output in Step 3;

[0062] Step 5: Construct a total loss function and joint training.

[0063] The following details the specific implementation processes of each step.

[0064] Step 1: Generate a batch of original CMYK images with a size of 1054×884 pixels, which conform to the A4 paper ratio, and each channel (cyan, magenta, yellow, black) maintains a uniform color output; output the generated original CMYK images through an inkjet printer, and use a scanner to obtain the printed RGB images; perform a color space conversion from the scanned RGB images to the XYZ space, and calculate the optical density (OD) values of each color.

[0065] Among them, the formula for converting the color space of the RGB image to the XYZ space is as follows:

[0066] X = 0.4124R + 0.3576G + 0.1805B

[0067] Y = 0.2126R + 0.7152G + 0.0722B

[0068] Z = 0.0193R + 0.1192G + 0.9505B

[0069] The formula for calculating the OD value of each color is as follows:

[0070] Cyan feature:

[0071] Magenta feature:

[0072] Yellow feature:

[0073] The original CMYK images generated in this step usually have their image content designed as a single color area with uniform color, such as a whole flat image of pure cyan or other single colors, and may also contain uniform color blocks of different colors. This batch of images undergoes strict color management and calibration to ensure that when sent to the printer, the values of each color channel are uniform and accurate, so that when measuring the printed output through scanning and spectrophotometer, the distortion generated by the printer can be accurately evaluated.

[0074] Step 2: Construct a reverse compensation network RCN, as Figure 2 shown. This network adopts a U-Net architecture, including an encoder, a bottleneck layer, and a decoder. The input is the original CMYK image, and the output is the compensated CMYK image; the encoder part contains multiple convolutional blocks, each convolutional block contains two layers of convolution and activation functions, followed by a pooling layer; the bottleneck layer contains multiple residual blocks; the decoder part gradually restores the spatial dimension through upsampling and skip connections, and finally outputs the compensated CMYK image through a convolutional layer.

[0075] In this embodiment, the input size of the input layer is a CMYK image of 1054×884×4; the encoder contains 4 convolutional blocks, each block consists of 2 layers of 3×3 convolution (the filters are 64, 128, 256, 512 respectively) and ReLU activation, followed by 2×2 max pooling; the bottleneck layer: adopts 3 residual blocks, each block consists of 2 layers of 3×3 convolution (512 filters), normalization and ReLU, and finally realizes residual connection; the decoder gradually restores the spatial size by upsampling and performs skip connection with the corresponding layer of the encoder; finally, a 4-channel compensated CMYK image is output through 1×1 convolution.

[0076] Step 3: Construct a printer distortion simulation network PDM, as Figure 3 shown. This network adopts a U-Net architecture similar to RCN, including a network input end, an encoder, a bottleneck layer and a decoder. The input is the compensated CMYK image, and the output is an RGB image after simulating the printing and scanning processes; a convolutional layer is set at the network input end to convert the CMYK image into an internal feature (color distribution and channel correlation, local texture and structure, non-linear distortion mode) representation; the designs of the encoder and the bottleneck layer are similar to those of RCN, but the number of filters in the encoder and the bottleneck layer is reduced; the decoder part finally outputs an RGB image through upsampling and skip connection.

[0077] In this embodiment, the input of the input layer is the 1054×884×4 compensated CMYK image output by RCN; a 1×1 convolutional layer is added to the input conversion layer to realize color space conversion (preliminary mapping from CMYK to RGB); the encoder is similar to the RCN encoder design, but the number of filters is slightly reduced, and a feature map is obtained after passing through 3 convolutional blocks; the bottleneck layer uses 2 residual blocks for feature extraction; the decoder gradually upsamples and fuses the encoder features, and finally outputs a 3-channel RGB image through 1×1 convolution.

[0078] Step 4: Perform regional processing on the RGB image output by PDM, calculate the color mean and standard deviation of each region, statistically calculate the macro uniformity score of the entire image (the calculation method of the macro uniformity score refers to the ISO / TS18621-21 standard), and construct a macro uniformity loss

[0079] Specifically, the macro uniformity loss is defined as follows: The image is divided into K regions (for example, 8×8 regions), and the average color value of each region is calculated; let μ i be the average color value of the i-th region, and be the average of the means of all regions, then the calculation formula of the macro uniformity loss is:

[0080]

[0081] After the scanned image is subjected to XYZ conversion, the OD values of cyan, magenta, and yellow are calculated respectively, denoted as to obtain the true OD C OD M OD Y diagram; perform the same processing on the RGB image output by PDM, that is, calculate the OD values of cyan, magenta, and yellow respectively after XYZ conversion, denoted as to obtain the predicted OD C OD M OD Y diagram; construct three OD losses respectively:

[0082] Cyan loss:

[0083] Magenta loss:

[0084] Yellow loss:

[0085] Step Five: Construct the total loss function Use the following combined loss function for training:

[0086]

[0087] where is the pixel-level loss; is the structural similarity loss; is the perceptual loss; is the macroscopic uniformity loss; are the OD losses of the cyan, magenta, and yellow features respectively; λ SSIM λ perc λ uin are the weights of each loss term;

[0088]

[0089] where I pred (i), I true (i) represent the pixel values of the predicted image and the true image at pixel i respectively, and N is the total number of pixels in the image;

[0090]

[0091] where the structural similarity (SSIM) index measures the similarity of images in terms of brightness, contrast, and structure, and its formula is:

[0092]

[0093] where μ x ​, μ y respectively represent the mean intensity values of each pixel of the predicted image and the ground truth image in the defined area, and is the local variance, σ xy is the covariance, and C1 and C2 are constants set to avoid a zero denominator;

[0094]

[0095] Among them, respectively represent the feature maps extracted by the pre-trained networks of the predicted image and the ground truth image at the l-th layer.

[0096] Through end-to-end joint training, the entire network (including RCN and PDM) is regarded as a whole. The overall loss is calculated through one forward propagation, and then the gradients of the two parts are calculated simultaneously through the backpropagation algorithm. Then, the gradient descent method is used to synchronously update all parameters. From the Figure 4 , Figure 5 comparison, it can be seen that after color compensation optimization, the inkjet printed images are optimized at all levels such as pixel level, structure level, perception level, macroscopic uniformity, and color characteristics. As Figure 6 shown, through the comparison of the macroscopic uniformity scores of the printed images before and after optimization with different color concentrations, the inkjet printed images have been significantly improved in terms of macroscopic uniformity.

[0097] Corresponding to the foregoing embodiments of the method for compensating the color uniformity of inkjet printed images based on deep learning, the present invention also provides embodiments of an apparatus for compensating the color uniformity of inkjet printed images based on deep learning.

[0098] Referring to Figure 7 , an apparatus for compensating the color uniformity of inkjet printed images based on deep learning provided by an embodiment of the present invention includes a memory and one or more processors. Executable code is stored in the memory. When the processor executes the executable code, it is used to implement the method for compensating the color uniformity of inkjet printed images based on deep learning in the foregoing embodiments.

[0099] The embodiments of the apparatus for compensating the color uniformity of inkjet printed images based on deep learning of the present invention can be applied to any device with data processing capabilities. The any device with data processing capabilities can be a device or apparatus such as a computer. The apparatus embodiments can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a logically meaningful apparatus, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions in the non-volatile memory into the memory for running. From the hardware level, as Figure 7As shown, it is a hardware structure diagram of any device with data processing capabilities where the inkjet printing image color uniformity compensation device based on deep learning of the present invention is located. Except for Figure 7 the shown processor, memory, network interface, and non-volatile memory, any device with data processing capabilities where the device in the embodiment is located usually may further include other hardware according to the actual functions of the any device with data processing capabilities, which will not be elaborated herein. The implementation processes of the functions and roles of each unit in the above device are specifically detailed in the implementation processes of the corresponding steps in the above method, which will not be elaborated herein.

[0100] An embodiment of the present invention further provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, the method for compensating the color uniformity of an inkjet printing image based on deep learning in the above embodiment is implemented.

[0101] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium may also be an external storage device of any device with data processing capabilities, such as a plug-in hard disk, a Smart Media Card (SMC), an SD card, a Flash Card, etc. equipped on the device. Further, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or will be output.

[0102] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, commodity or device including the said element.

[0103] The specific embodiments of this specification have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0104] The above are only the preferred embodiments of one or more embodiments of this specification, and are not intended to limit one or more embodiments of this specification. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of one or more embodiments of this specification shall be included within the scope of protection of one or more embodiments of this specification.

[0105] The above is only the specific implementation manner of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or replacements, which should all be covered within the scope of protection of the present invention.

Claims

1. An inkjet printing image color uniformity compensation method based on deep learning, characterized in that, It includes the following steps: (1) Data acquisition and preprocessing: Generate the original CMYK image, and each channel maintains a uniform color output; Output the generated original CMYK image through an inkjet printer, and use a scanner to obtain the printed RGB image; Convert the scanned RGB image from the color space to the XYZ space, and calculate the optical density OD value of each color feature; (2) Construct the reverse compensation network RCN, input the original CMYK image into RCN, and output the compensated CMYK image; (3) Construct the printer distortion simulation network PDM, input the compensated CMYK image into PDM, and output the RGB image to simulate the non-linear color distortion generated during the printing and scanning processes; (4) Conduct a macroscopic uniformity detection on the RGB image output in step (3); (5) Construct a total loss function including pixel-level loss, structural similarity loss, perceptual loss, macroscopic uniformity loss, and OD loss of different color features and conduct joint training, finally optimizing the inkjet printed image at the pixel level, structural level, perceptual level, macroscopic uniformity, and color feature levels.

2. The inkjet printing image color uniformity compensation method according to claim 1, wherein, (1) Specifically, generate a batch of original CMYK images with a size of 1054×884 pixels, conforming to the A4 paper ratio, and the cyan, magenta, yellow, and black channels all maintain a uniform color output; Output the generated original CMYK image through an inkjet printer, and use a scanner to obtain the printed RGB image; Convert the scanned RGB image from the color space to the XYZ space, and calculate the optical density OD value of each color; The formula for converting the color space of the RGB image to the XYZ space is as follows: X = 0.4124R + 0.3576G + 0.1805B Y = 0.2126R + 0.7152G + 0.0722B Z = 0.0193R + 0.1192G + 0.9505B The formula for calculating the OD value of each color is as follows: Cyan feature: Magenta feature: Yellow feature:

3. The inkjet printing image color uniformity compensation method according to claim 1, characterized in that (2) Specifically, construct the reverse compensation network RCN. This network adopts the U-Net architecture, including an encoder, a bottleneck layer, and a decoder. The input is the original CMYK image, and the output is the compensated CMYK image; The encoder part contains multiple convolutional blocks, each convolutional block contains two layers of convolution and activation functions, followed by a pooling layer; The bottleneck layer contains multiple residual blocks; The decoder part gradually restores the spatial dimension through upsampling and skip connections, and finally outputs the compensated CMYK image through a convolutional layer.

4. The method for compensating color uniformity of an inkjet printed image according to claim 1, characterized in that, (3) Specifically, construct the printer distortion simulation network PDM. This network adopts the U-Net architecture, including a network input end, an encoder, a bottleneck layer, and a decoder. The input is the compensated CMYK image, and the output is the RGB image after simulating the printing and scanning processes; A convolutional layer is set at the network input end to convert the CMYK image into an internal feature representation; The encoder part contains multiple convolutional blocks, each convolutional block contains two layers of convolution and activation functions, followed by a pooling layer; The bottleneck layer contains multiple residual blocks; The decoder part outputs the RGB image through upsampling and skip connections.

5. The inkjet printing image color uniformity compensation method according to claim 1, characterized in that Step (4) specifically is: perform regional processing on the RGB image output by PDM, calculate the color mean and standard deviation of each region, statistically calculate the macroscopic uniformity score of the entire image, and construct a macroscopic uniformity loss The macroscopic uniformity loss is defined as follows: divide the image into K regions and calculate the average color value of each region; let μ i be the average color value of the i-th region, and be the average of the means of all regions, then the calculation formula for the macroscopic uniformity loss is: After the scanned image is subjected to XYZ conversion, the OD values of cyan, magenta, and yellow are calculated respectively, denoted as to obtain the true OD C , OD M , OD Y graph; The same processing is performed on the RGB image output by the PDM, that is, after XYZ conversion, the OD values of cyan, magenta, and yellow are calculated respectively, denoted as to obtain the predicted OD C , OD M , OD Y graph; Three OD losses are constructed respectively: Cyan loss: Magenta loss: Yellow loss:

6. The inkjet printing image color uniformity compensation method according to claim 5, characterized in that, Step (5) is specifically: constructing the total loss function The following combined loss function is used for training: Among them, is the pixel-level loss; is the structural similarity loss; is the perceptual loss; is the macroscopic uniformity loss; are the OD losses of the cyan, magenta, and yellow features respectively; λ SSIM , λ perc , λ uin , are the weights of each loss term.

7. The method for compensating the color uniformity of an inkjet printed image according to claim 6, wherein The calculation formula of the pixel-level loss is: where, i pred (i), I true (i) respectively represent the pixel values of the predicted image and the ground truth image at pixel i, and N is the total number of pixels in the image; The calculation formula of the structural similarity loss is as follows: Among them, the structural similarity SSIM index measures the similarity of images in terms of brightness, contrast, and structure, and its formula is: Among them, μ x and μ y respectively represent the mean intensity values of each pixel in the defined area of the predicted image and the real image. and are the local variances, σ xy is the covariance, and C1 and C2 are constants set to avoid a zero denominator; The calculation formula of the perceptual loss is as follows: Among them, respectively represent the feature maps extracted by the pre-trained networks of the predicted image and the real image at the l-th layer.

8. The inkjet printing image color uniformity compensation method according to claim 1, characterized in that In step (5), through end-to-end joint training, the entire network composed of RCN and PDM is regarded as a whole. The overall loss is calculated through one forward propagation, and then the gradients of the two parts are calculated simultaneously through backpropagation. Then, the gradient descent method is used to synchronously update all parameters, and finally, the inkjet printed image is optimized at all levels such as pixel level, structural level, perceptual level, macroscopic uniformity, and color characteristics.

9. An inkjet printing image color uniformity compensation device based on deep learning, comprising a memory and one or more processors, wherein executable code is stored in the memory, characterized in that, When the processor executes the executable code, it is used to implement the method for compensating the color uniformity of inkjet printed images based on deep learning according to any one of claims 1-8.

10. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it is used to implement the method for compensating the color uniformity of inkjet printed images based on deep learning according to any one of claims 1-8.

Images