Color table extraction neural network model construction method and system and color table extraction method and system

Abstract

The invention provides a color table extraction neural network model construction method and system and a color table extraction method and system, and belongs to the technical field of computer image processing and data visualization. Inputting the single scalar field visualization image into a pre-trained color table extraction neural network model, wherein the color table extraction neural network model processes the input image to extract scalar data and a color control point sequence; interpolating a color control point sequence to generate a continuous color table, and mapping the continuous color table and the extracted scalar data to generate a reconstructed image; self-supervised fine tuning is carried out in the reasoning stage of the color table extraction neural network model, and a final continuous color table is extracted.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer image processing and data visualization, and in particular relates to the construction of a color table extraction neural network model, a color table extraction method and a system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] A continuous colormap is a color mapping scheme used to visualize continuous data. It maps a range of values to a series of smoothly transitioning colors. In scientific visualization and image analysis, continuous colormaps are widely used to visualize scalar field data. By mapping different colors to different data values, they make the spatial distribution characteristics and numerical variations of the data more intuitive.

[0004] However, in reality, a large number of visual images (such as academic publications, web page screenshots, report charts, etc.) lack clear legends or color map information, which makes the data information in these images difficult for machines or humans to accurately understand, reuse or redesign.

[0005] Among them, the legend is a key component in data visualization that explains the meaning of graphic elements. Legend design can significantly improve the readability of charts. Existing color map extraction methods mostly rely on legends to map colors to data.

[0006] Existing techniques for color table extraction often rely on legends to map colors to data. This often involves combining OCR (Optical Character Recognition) and image processing techniques to obtain the color-value pairs in the legend. These pairs are crucial elements in visualizations for explaining the correspondence between color codes and data values. However, in the absence of a legend, these methods are ineffective, and the extracted color map lacks accuracy and interpretability.

[0007] In addition, some supervised learning methods rely on paired training samples (image and color map pairs) for model training, which has weak generalization ability and is difficult to adapt to real-world scenarios with diverse image distributions in different fields. Summary of the Invention

[0008] To overcome the deficiencies of the above-mentioned prior art, the present invention provides a self-supervised color table extraction method and system for scalar data visualization images. This method can accurately recover the continuous color map and underlying scalar data used in a single visualization image without the need for paired color map information or legends, thereby improving the analysis and reuse capabilities of visualization images.

[0009] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: In a first aspect, a method for constructing a color table extraction neural network model is disclosed, comprising: Preprocess a single scalar field visualization image into an image in a standard format; Input the image preprocessed into a standard format into the visual image encoder to extract the potential deep features of the image; The extracted deep features are divided into two branches for processing: one branch generates grayscale scalar data corresponding to the image, and the other branch generates a color control point sequence through a multi-layer perceptron network. The color control point sequence is then interpolated through spline to generate a continuous color table. Perform color mapping on the extracted scalar data and the continuous color table to generate a reconstructed image; The reconstructed image is compared with the original single scalar field visualization image, and the continuous color table and scalar data are compared with the real continuous color table and real data. Multiple self-supervised loss functions are calculated respectively, and the neural network parameters and color control points are continuously updated through backpropagation to perform self-supervised fine-tuning and obtain the color table extraction neural network model.

[0010] As a further technical solution, the color table extraction neural network model includes: Visual image encoder, data decoder and color decoder; The visual image encoder is used to extract the latent representation of an input single scalar field visual image; The data decoder restores the image into scalar data based on the latent representation of the image; The color decoder outputs a fixed number of color control point sequences, and constructs a continuous color table through spline interpolation.

[0011] As a further technical solution, self-supervised fine-tuning is performed through multiple self-supervised loss functions. The loss functions specifically include: Reconstruction loss function and color order loss function between the input image and the reconstructed image; Data loss function and color loss function between the scalar data and continuous color table generated in the initial iteration and the scalar data and continuous color table generated in each subsequent iteration.

[0012] As a further technical solution, the reconstruction loss function between the input image and the reconstructed image is used to measure the pixel difference between the input image and the reconstructed image.

[0013] As a further technical solution, the color order loss function between the input image and the reconstructed image is used to penalize color order anomalies between color control point sequences.

[0014] As a further technical solution, a data loss function between the scalar data and continuous color table generated in the initial iteration and the scalar data and continuous color table generated in each subsequent iteration is used to measure the difference between the extracted scalar data and the real data.

[0015] As a further technical solution, a color loss function between the scalar data and continuous color table generated in the initial iteration and the scalar data and continuous color table generated in each subsequent iteration is used to measure the difference between the extracted color table and the original color table.

[0016] In a second aspect, a color table extraction method is disclosed, comprising: Obtain a single scalar field visualization image to be processed; Inputting a single scalar field visualization image into a pre-trained color table extraction neural network model, wherein the color table extraction neural network model processes the input image to extract scalar data and a color control point sequence; interpolating a sequence of color control points to generate a continuous color table, and then mapping the continuous color table and the extracted scalar data to generate a reconstructed image; Self-supervised fine-tuning is performed during the inference phase of the color table extraction neural network model, including obtaining the difference between the input single scalar field visualization image and the reconstructed image, and iteratively optimizing the color control point sequence and the network parameters of the related color table extraction neural network model through a back-propagation mechanism based on the obtained difference to extract the final continuous color table.

[0017] In a third aspect, a color table extraction system is disclosed, comprising: The image acquisition module is configured to: acquire a single scalar field visualization image; An image processing module is configured to: input a single scalar field visualization image into a pre-trained color table extraction neural network model, wherein the color table extraction neural network model processes the input image to extract scalar data and a color control point sequence; a reconstructed image generation module configured to: interpolate a color control point sequence to generate a continuous color table, and then map the continuous color table with the extracted scalar data to generate a reconstructed image; The self-supervised fine-tuning module is configured to perform self-supervised fine-tuning during the inference phase of the color table extraction neural network model, including obtaining the difference between the input single scalar field visualization image and the reconstructed image, and iteratively optimizing the color control point sequence and the network parameters of the related color table extraction neural network model through a back-propagation mechanism based on the obtained difference to extract the final continuous color table.

[0018] One or more of the above technical solutions have the following beneficial effects: In the technical solution of the present invention, during the model training process, the difference between the input single scalar field visualization image and the reconstructed image is first obtained. Based on this difference, the network parameters of the color control point sequence and the related color table extraction neural network model are iteratively optimized through a backpropagation mechanism to extract the final continuous color table. This method optimizes the network during the inference phase through a self-supervised fine-tuning mechanism, enabling the system to adapt to images with large distribution differences, significantly improving the model's robustness and generalization ability in real complex scenes. This method can be applied not only to color table inversion and image recoloring, but also to color table migration technology.

[0019] During the inference phase of the color table extraction neural network model, the present invention employs multiple self-supervised loss functions to perform self-supervised fine-tuning. These loss functions incorporate multiple constraints, including color fidelity loss, data reconstruction loss, and color order loss, effectively improving the accuracy and visual consistency of color table extraction. In particular, the use of a control point-based spline interpolation method enables the generation of a continuous color table with smooth transitions and consistent order, avoiding the color band jumps associated with traditional linear interpolation.

[0020] The technical solution of the present invention inputs a single scalar field visualization image into a pretrained color table extraction neural network model. The color table extraction neural network model processes the input image to extract scalar data and a sequence of color control points. This approach accurately extracts the color table and implicit scalar data used in a single visualization image, and is applicable to both legend-free and legend-free scenarios. The color control point sequence is then interpolated to generate a continuous color table, which is then mapped to the extracted scalar data to generate a reconstructed image. This effectively separates and reconstructs the color and data information in the image, improving the interpretability, usability, and automatic processing capabilities of the visualization image.

[0021] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0023] Figure 1 Schematic diagram of the process of a self-supervised color table extraction method for scalar field visualization images according to an embodiment of the present invention; Figure 2 This is an architecture diagram of a color extraction neural network according to an embodiment of the present invention; Figure 3Schematic diagram of qualitative results of an embodiment of the present invention on a synthetic dataset and a real-world dataset. DETAILED DESCRIPTION

[0024] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0025] It should be noted that the terms used herein are for describing particular embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0026] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0027] Example 1 This embodiment discloses a method for constructing a color table extraction neural network model, including: Preprocess a single scalar field visualization image into an image in a standard format; Input the image preprocessed into a standard format into the visual image encoder to extract the potential deep features of the image; The extracted deep features are divided into two branches for processing: one branch generates grayscale scalar data corresponding to the image, and the other branch generates a color control point sequence through a multi-layer perceptron network. The color control point sequence is then interpolated through spline to generate a continuous color table. Perform color mapping on the extracted scalar data and the continuous color table to generate a reconstructed image; The reconstructed image is compared with the original single scalar field visualization image, and the continuous color table and scalar data are compared with the real continuous color table and real data. Multiple self-supervised loss functions are calculated respectively, and the neural network parameters and color control points are continuously updated through backpropagation to perform self-supervised fine-tuning and obtain the color table extraction neural network model.

[0028] More specifically, the color table extraction neural network model is based on the training of the color extraction neural network. The specific training process is as follows: Taking the training process of a single image as an example, the model inputs a scalar field visualization image I and a neural network N. The image is first scaled to a uniform size and normalized to a standard format, then fed into a visualization image encoder to extract latent deep features z. The extracted deep features are then processed in two branches: one path generates grayscale scalar data D(I) corresponding to the image through a U-Net architecture, and the other path generates vectors P(I) representing color control points through a multilayer perceptron network. These control points are interpolated using cubic B-spline interpolation to generate a continuous color table C(I). The model then performs color mapping using the extracted scalar data and the continuous color table, generating a reconstructed image R(I). The model then compares the reconstructed image with the original image and compares C(I) and D(I) with the ground-truth continuous color table and the real data. Multiple self-supervised loss functions, including image reconstruction loss, color order loss, data loss, and color loss, are calculated. Backpropagation is then used to continuously update the neural network parameters and color control points for self-supervised fine-tuning, gradually improving the model's extraction capabilities and output quality.

[0029] To construct this training data, we first collected two types of 2D scalar field data: 179 meteorological data points from the ECMWF dataset and 228 scalar field data points generated using various mathematical functions (such as linear gradients). Furthermore, we collected 179 color tables from the Python matplotlib library and academic papers. Based on this collected scalar field data and color tables, we used linear mapping to generate corresponding scalar field visualization images, totaling 72,853 visualization images. The training set includes scalar field visualization images, color tables, and scalar field data.

[0030] In order to make the above model perform better on the input image processing performance, in the inference stage, self-supervised fine-tuning is done by visualizing the reconstruction loss and color order loss between the input scalar field image I and the reconstructed image R(I). The scalar data generated in the initial iteration is and the continuous color table and the scalar data generated by each subsequent round (i is the number of iterations) and continuous color tables The data loss and color loss between the multiple self-supervised loss functions are continuously updated through back propagation, and self-supervised fine-tuning is performed to gradually improve the extraction ability and output quality of the model.

[0031] More specifically, the loss functions include the following four, which are used for joint training and self-supervised fine-tuning: Image reconstruction loss , which is used to measure the pixel difference between the scalar field visualization image and the reconstructed scalar field visualization image, is defined using the mean square error (MSE) formula as:

[0032] in, W and H are the width and height of the input scalar field visualization, respectively, is the original scalar field visualization image in the two-dimensional image space The color value at Is to reconstruct the scalar field visualization in two-dimensional image space The color value at .

[0033] Data loss , which is used to measure the difference between the extracted scalar data and the real data. The formula is defined as:

[0034] in, is the normalized scalar value (ranging from 0 to 1) of the original scalar field data in the two-dimensional data space (h, w), is the normalized scalar value of the extracted scalar field data in the two-dimensional data space (h, w), W and H are the width and height of the input scalar field data, respectively, which are consistent with the width and height of the scalar field visualization image.

[0035] Color loss , used to measure the difference between the extracted color table and the true color table, the formula is defined as:

[0036] in, is the i-th color in the original colormap, is the color at parameter position t(i) in the extracted colormap, and m is the number of colors in the colormap.

[0037] Color order loss , used to penalize color sorting anomalies between control points, the formula is defined as:

[0038] in, and is the color value at parameter position t(i) and t(j) in the extracted color table, where i and j are the indices of the color, ranging from 1 to m.

[0039] The image reconstruction loss function evaluates the quality of color map extraction from a global perspective by remapping the extracted color map and data into an image and comparing it to the original scalar field image. This loss function provides an end-to-end optimization objective to ensure overall consistency and can also serve as a self-supervisory signal in unsupervised or semi-supervised scenarios to improve network generalization.

[0040] The data loss function is mainly used to measure the difference between scalar field data and real data. This loss function helps to improve the restoration accuracy of scalar field data, thereby strengthening the consistency of color and data decoupling. The color loss function mainly measures the difference between the predicted color table and the true color table, which helps to improve the extraction accuracy of the color table and ensure that the predicted color table is as consistent as possible with the true color table.

[0041] The color order loss function primarily ensures that the colors in the recovered colormap are arranged in a certain order that is consistent with human perception. This loss function ensures the coherence and interpretability of color changes, avoiding perceptual ambiguity caused by color disorder. This loss function is particularly effective when the input visualization image lacks a legend (color map).

[0042] During the model training and inference phases, the above losses are weighted proportionally to form a total loss function, which is defined as:

[0043] in, 、 、 It is an adjustable weight coefficient, which is set to 1 by default.

[0044] The advantage of the above training scheme is that it not only constructs an end-to-end color table extraction solution but also improves the generalization ability of the color table extraction method through self-supervised fine-tuning by simultaneously decoupling the color table information and grayscale scalar data information in the scalar field visualization image, fusing B-spline curves and color table order constraints, and introducing a differentiable color table mapping module.

[0045] In another embodiment, a color table extraction method is disclosed, comprising: Step 1: Obtain a single scalar field visualization image; Step 2: Input the image into the pre-trained color table extraction neural network model to obtain the extracted scalar data and color control point sequence; Step 3: The control point sequence output by the color table decoder is interpolated through the cubic B-spline function to generate a continuous color table curve, and the reconstructed image is generated through the differentiable color table mapping module with the obtained scalar data; Step 4: Perform self-supervised fine-tuning during the model inference phase. Based on the difference between the input image and the reconstructed image, the control points and related network parameters are iteratively optimized through the back-propagation mechanism to extract a continuous color table. Among them, the color table extraction neural network model, its training process defines and jointly uses multiple loss functions, including reconstruction loss function, data loss function, color loss function and color sorting loss function, to optimize the model.

[0046] In the technical solution of this example embodiment, the color table extraction neural network model processes the input image to extract scalar data and a sequence of color control points. This approach accurately extracts the color table and implicit scalar data used in a single visual image, applicable to both the presence and absence of legends. The color control point sequence is then interpolated to generate a continuous color table, which is then mapped to the extracted scalar data to generate a reconstructed image. This effectively separates and reconstructs the color and data information in the image, improving the interpretability, usability, and automated processing capabilities of the visual image. This addresses the existing inability to accurately extract the color table and underlying data in the absence of legend information.

[0047] In step 1, a single scalar field visualization image to be processed is obtained, and the obtained visualization image is preprocessed to set the image size to 256×256 as the input image of the neural network model.

[0048] In one embodiment, step 2 inputs the image into a pre-trained color table extraction neural network model to obtain the extracted scalar data and color control point sequence. Figure 2 As shown, the model consists of a visual image encoder, a data decoder, and a color decoder. The visual image encoder consists of one convolutional layer and six downsampling modules. Each module consists of one residual block and one downsampling (stride = 2). The residual block contains two 3×3 convolutional layers. The visual image encoder is used to extract the latent representation of the input image. The data decoder adopts a U-Net structure and consists of six upsampling modules with skip connections. Each upsampling module contains a nearest neighbor upsampling (scale = 2) and a residual block. The residual block contains two 3×3 convolutions. The skip connections are symmetrical to the encoder layer. The data decoder decouples the input and restores the scalar data. The color decoder uses a multi-layer perceptron (MLP) with six fully connected neural networks. The color decoder outputs a fixed number of control points, which are used to construct a continuous color table through cubic B-spline interpolation. The above model can extract the color table and latent data from a single visual image.

[0049] In step three, a differentiable color table mapping module is specifically used to receive the scalar data and the continuous color table decoupled by the neural network decoder. The neural network decoder includes a data decoder and a color table decoder. The color values arranged in sequence in the continuous color table are mapped to the size of the grayscale values in the scalar data to obtain a reconstructed image.

[0050] The grayscale values in the scalar data output by the neural network are mapped one-to-one with the colors in the continuously differentiable color table, that is, the range of grayscale values is normalized. Each grayscale value corresponds to the color value of a position in the color table to realize the conversion from scalar to RGB color. The grayscale values from small to large correspond to the color values from left to right in the continuous color table.

[0051] The network is responsible for extracting features and outputting scalar data and color control points. The differentiable color mapping module takes the output of the main network as input and completes the final color image reconstruction. The reconstructed image is compared with the original image and a loss (such as MSE) is calculated. This loss is then back-propagated to the color table and the main network. The main training loop calls the main network and the color mapping module to obtain the final output and calculate the loss. This differentiable color table mapping module converts traditional discrete color mapping into a differentiable continuous operation, allowing the entire color conversion process to be learned and optimized by a neural network, achieving an end-to-end trainable mapping from scalar data to color images.

[0052] In order to prove the effectiveness of the solution of the present invention, specific experiments were carried out in this example to prove that: Table 1 shows a quantitative comparison of the method in this example with the baseline method in RGB space on both synthetic and real-world datasets. Three different evaluation metrics were used, including image reconstruction error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity (SSIM), including whether or not the color table direction was considered. It can be seen that the method in this example outperforms the baseline method. This result demonstrates that in terms of image reconstruction capabilities, the method in this example comprehensively outperforms the traditional baseline method in terms of MSE, PSNR, and SSIM metrics, indicating that the model can more accurately restore the color distribution and structural information of the input image.

[0053] Table 1

[0054] exist Figure 3In this paper, we present a qualitative comparison of the color map extraction results of the method in this example and a baseline method on synthetic and real-world datasets. Each dataset shows six examples with corresponding ground-truth values. Input is the input visualization image, GT is the ground-truth color map, Ours is the method in this example, and Yuan et al. is the baseline method. As can be seen, the results of the method in this example on both datasets are closer to the ground-truth than those of the baseline method, and are able to extract higher-quality color maps.

[0055] Example 2 The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the program.

[0056] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0057] A computer-readable storage medium stores a computer program, which, when executed by a processor, performs the steps of the above method.

[0058] Example 4 The purpose of this embodiment is to provide a self-supervised color table extraction system for scalar data visualization images, including: The image acquisition module is configured to: acquire a single scalar field visualization image; An image processing module is configured to: input a single scalar field visualization image into a pre-trained color table extraction neural network model, wherein the color table extraction neural network model processes the input image to extract scalar data and a color control point sequence; a reconstructed image generation module configured to: interpolate a color control point sequence to generate a continuous color table, and then map the continuous color table with the extracted scalar data to generate a reconstructed image; The self-supervised fine-tuning module is configured to perform self-supervised fine-tuning during the inference phase of the color table extraction neural network model, including obtaining the difference between the input single scalar field visualization image and the reconstructed image, and iteratively optimizing the color control point sequence and the network parameters of the related color table extraction neural network model through a back-propagation mechanism based on the obtained difference to extract the final continuous color table.

[0059] Example 5 The purpose of this embodiment is to provide a computer program product containing instructions, which, when running on a computer, enables the computer to execute the methods and functions involved in any of the above embodiments. The steps involved in the apparatus of the above embodiment correspond to those of the method embodiment 1. For detailed implementation, please refer to the relevant description of embodiment 1. The term "computer-readable storage medium" should be understood to mean a single medium or multiple media containing one or more instruction sets; it should also be understood to include any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and causing the processor to perform any method of the present invention.

[0060] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0061] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A method for constructing a color table extraction neural network model, characterized by: include: Preprocess a single scalar field visualization image into an image in a standard format; Input the image preprocessed into a standard format into the visual image encoder to extract the potential deep features of the image; The extracted deep features are divided into two branches for processing: one branch generates grayscale scalar data corresponding to the image, and the other branch generates a color control point sequence through a multi-layer perceptron network. The color control point sequence is then interpolated through spline to generate a continuous color table. Perform color mapping on the extracted scalar data and the continuous color table to generate a reconstructed image; The reconstructed image is compared with the original single scalar field visualization image, and the continuous color table and scalar data are compared with the real continuous color table and real data. Multiple self-supervised loss functions are calculated respectively, and the neural network parameters and color control points are continuously updated through backpropagation to perform self-supervised fine-tuning and obtain the color table extraction neural network model.

2. The color table extraction neural network model construction method according to claim 1, wherein: The color table extraction neural network model includes: Visual image encoder, data decoder and color decoder; The visual image encoder is used to extract the latent representation of an input single scalar field visual image; The data decoder restores the image into scalar data based on the latent representation of the image; The color decoder outputs a fixed number of color control point sequences and constructs a continuous color table through spline interpolation.

3. The color table extraction neural network model construction method according to claim 1, wherein: Self-supervised fine-tuning is performed through multiple self-supervised loss functions, including: Reconstruction loss function and color order loss function between the input image and the reconstructed image; Data loss function and color loss function between the scalar data and continuous color table generated in the initial iteration and the scalar data and continuous color table generated in each subsequent iteration.

4. The color table extraction neural network model construction method according to claim 1, wherein: The reconstruction loss function between the input image and the reconstructed image is used to measure the pixel difference between the input image and the reconstructed image; The color order loss function between the input image and the reconstructed image is used to penalize color order anomalies between color control point sequences.

5. The color table extraction neural network model construction method according to claim 1, wherein: A data loss function between the scalar data and continuous color table generated in the initial iteration and the scalar data and continuous color table generated in each subsequent iteration is used to measure the difference between the extracted scalar data and the true data; The color loss function between the scalar data and continuous color table generated in the initial iteration and the scalar data and continuous color table generated in each subsequent iteration is used to measure the difference between the extracted color table and the original color table.

6. A color table extraction method, characterized in that: The color table extraction neural network model obtained by the color table extraction neural network model construction method according to any one of claims 1 to 5 is implemented, and the method comprises: Obtain a single scalar field visualization image to be processed; Inputting a single scalar field visualization image into a pre-trained color table extraction neural network model, wherein the color table extraction neural network model processes the input image to extract scalar data and a color control point sequence; interpolating a sequence of color control points to generate a continuous color table, and then mapping the continuous color table and the extracted scalar data to generate a reconstructed image; Self-supervised fine-tuning is performed during the inference phase of the color table extraction neural network model, including obtaining the difference between the input single scalar field visualization image and the reconstructed image, and iteratively optimizing the color control point sequence and the network parameters of the related color table extraction neural network model through a back-propagation mechanism based on the obtained difference to extract the final continuous color table.

7. A color table extraction system, characterized in that: The color table extraction neural network model obtained based on the color table extraction neural network model construction method according to any one of claims 1 to 5 is implemented, and the system includes: The image acquisition module is configured to: acquire a single scalar field visualization image; An image processing module is configured to: input a single scalar field visualization image into a pre-trained color table extraction neural network model, wherein the color table extraction neural network model processes the input image to extract scalar data and a color control point sequence; a reconstructed image generation module configured to: interpolate a color control point sequence to generate a continuous color table, and then map the continuous color table with the extracted scalar data to generate a reconstructed image; The self-supervised fine-tuning module is configured to perform self-supervised fine-tuning during the inference phase of the color table extraction neural network model, including obtaining the difference between the input single scalar field visualization image and the reconstructed image, and iteratively optimizing the color control point sequence and the network parameters of the related color table extraction neural network model through a back-propagation mechanism based on the obtained difference to extract the final continuous color table.

8. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the method described in any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are performed.

Images