Intelligent color mapping method and device based on feature saliency, equipment and storage medium

Abstract

The invention discloses an intelligent color mapping method and device based on feature saliency, equipment and a storage medium, and relates to the technical field of scientific visualization, and the method comprises the steps: carrying out the clustering of pixel points of a target style image, obtaining a plurality of control points, determining a color sequence adjustment function of the control points based on the perception uniformity deviation and the smoothing error, and obtaining a color sequence adjustment function of the target style image; performing color mapping on each control point by using a color mapping table generated by the color sequence adjustment function to obtain a mapping image, acquiring a feature image for identifying the feature region to be highlighted, and processing the mapping image by using an MSRMNet network model to obtain a saliency image; processing the mapping image, the feature image and the saliency image based on the feature saliency item, the color pair project and the feature sensitivity item to obtain a loss function; and determining a control point adjustment scheme based on a position adjustment result obtained by adjusting the position of each control point by using an annealing algorithm and a loss function. Therefore, the efficiency of adjusting the color mapping control points can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of scientific visualization, and particularly to an intelligent color mapping method, device, equipment and storage medium based on feature saliency. Background Technique

[0002] Scientific visualization is a technical discipline in the field of research on how to use computer graphics and image processing technologies to convert data generated by scientific calculations and data obtained by scientific measurements into graphics or images for display on a screen and perform interactive processing. Among them, flow field visualization is an important research direction in scientific visualization and an indispensable important part of flow field scientific calculations, namely Computational Fluid Dynamics (CFD). At present, flow field visualization is playing an increasingly important role in analyzing and understanding the complex flow mechanism of the flow field, insight into the physical phenomena of the flow field, and discovery of flow science laws. For more than two decades, flow field visualization has been one of the directions with the most achievements in the research of technical methods and the development of application software tools in the field of scientific computing visualization. As the core tool of scientific visualization, color mapping can help researchers quickly identify and analyze key feature regions in data by mapping data values to the color space. Therefore, the design of color mapping directly affects the readability and analysis efficiency of data. However, the data types of flow field scientific calculations are complex, there are many types of physical quantities, it is difficult to extract physical features, and the visualization requirements are very high, making the expression of feature regions through color mapping always one of the research hotspots and difficulties in the field of flow field visualization, and still facing new challenges so far.

[0003] In a broad sense, a "feature region" refers to the region that researchers focus on in a data set and is of great significance for revealing the key characteristics and laws of the data. Researchers can determine these regions through feature extraction or interaction methods. In scientific visualization, color mapping usually consists of multiple color control points, and each control point has two attributes: color value and position. By adjusting the positions and color values of these control points, the mapping relationship from data to color can be realized. A reasonably designed color mapping can highlight the feature regions in the data, such as vortex regions, boundary regions or outlier regions.

[0004] However, current data-driven automated design methods have limitations in processing complex scientific simulation data: they usually treat all data points equally and apply a unified algorithm for feature expression and definition. However, this approach is difficult to adapt to the unevenness and high noise of data distribution, resulting in a lack of robustness in practical applications and the need for a large amount of adjustment or even possible failure in the face of different data types. In addition, existing methods often ignore the need to meet the subjective preferences of different users and lack flexible customization options.

[0005] Therefore, in order to better observe the feature region, existing tools construct color mapping relationships in a manual interaction manner, but interaction often means relying on experience and being time-consuming and boring. In order to reduce the interaction time and improve efficiency, many methods for automatically generating or adjusting color mapping tables have been proposed one after another. The most commonly used is the one-dimensional color map. Although 2D (Two-Dimensional) color maps have been proposed for visualizing bivariate or multivariate data, such as ColorMapND (ColorMap n-Dimensional), they must be used with caution because they may be difficult to interpret. Therefore, this paper focuses on introducing the one-dimensional color mapping table design technology:

[0006] 1. An interpolation method that better utilizes the entire color space by increasing the discrimination attributes between similar values and reducing the differences between different values;

[0007] 2. Based on the Knowledge-Augmented Vision (KAV) paradigm, a Support Vector Machine (SVM) model is trained with preference data collected through web user surveys to automatically predict the optimized hue-preserving blending result, aiming to avoid false colors generated when using standard Alpha Compositing in visualization;

[0008] 3. Identify the prominent values and ranges in the data through random sampling, and then adjust the colors using the method of histogram equalization;

[0009] 4. Consider the background color and the opacity change of the color scale when designing the color map to match people's inferred mapping;

[0010] 5. Adjust the color mapping to a non-linear constrained optimization problem and reveal the spatial variation by integrating the boundary model of Kindlmann (a scientific visualization technique for generating perceptually uniform color mappings). However, the above methods work well for cases with obvious boundaries, but it is difficult to clearly reveal features with insignificant data value changes;

[0011] 6. Enhance the perceptual discriminability of the color table based on segmented brightness parameters, and finally generate a color mapping with adaptive data distribution characteristics. However, segmented brightness may lead to a violation of color consistency, resulting in the same color representing different values and causing confusion.

[0012] As can be seen from the above, how to improve the efficiency of adjusting color mapping control points in the intelligent color mapping process based on feature saliency is an urgent problem to be solved at present. Summary of the Invention

[0013] In view of this, the purpose of the present invention is to provide an intelligent color mapping method, device, equipment and storage medium based on feature saliency, which can improve the efficiency of generating a color mapping control point adjustment scheme in the intelligent color mapping process based on feature saliency, and further improve the efficiency of adjusting color mapping control points. The specific scheme is as follows:

[0014] In the first aspect, the present application provides an intelligent color mapping method based on feature saliency, including:

[0015] Cluster the pixel points corresponding to the target style image to obtain a number of control points, and determine a color order adjustment function corresponding to the control points based on the perceptual uniformity deviation and the smoothing error, and then generate a color mapping table using the color order adjustment function; the perceptual uniformity deviation is the degree of deviation between the current perceptual distance and the preset ideal perceptual distance of each adjacent control point in the perceptually uniform color space; the smoothing error is the color change smoothness error determined by the angular change of the displacement vectors corresponding to each control point;

[0016] Perform color mapping on each control point using the color mapping table to obtain a mapped image, and obtain a feature image for identifying the feature region to be highlighted, and process the mapped image using the MSRMNet network model to obtain a saliency image;

[0017] Based on the feature saliency term, the color opposition term and the feature sensitivity term, process the mapped image, the feature image and the saliency image to obtain a loss function; the feature saliency term is the degree of overlap between the saliency image and the feature image; the color opposition term is the difference in the average color modulus between the feature region and the non-feature region in the feature image; the feature sensitivity term is the maximum distance of the colors in the feature region in the color space;

[0018] Use the annealing algorithm and the loss function to adjust the positions of each control point to obtain a position adjustment result, and determine a control point adjustment scheme corresponding to the target style image based on the position adjustment result.

[0019] Optionally, before determining the color sequence adjustment function corresponding to the control points based on the perceptual uniformity deviation and the smoothing error, and then generating a color mapping table using the color sequence adjustment function, the method further includes:

[0020] Determine control point pairs based on every two of the control points, obtain the current perceptual distances between the control point pairs, and then determine the average distance corresponding to each of the current perceptual distances;

[0021] Determine the distance differences between the perceptual distances and the corresponding average distances, perform square operations and summations on the distance differences corresponding to the control point pairs to obtain a summation result, and then determine a calculation result based on the summation result and the number of all the control point pairs;

[0022] Determine a perceptual uniformity deviation value based on the calculation result and the average distance; the numerical value of the perceptual uniformity deviation value is negatively correlated with the magnitude of the perceptual uniformity corresponding to the control points.

[0023] Optionally, before determining the color sequence adjustment function corresponding to the control points based on the perceptual uniformity deviation and the smoothing error, and then generating a color mapping table using the color sequence adjustment function, the method further includes:

[0024] Determine a first displacement vector between each of the control points and the previous control point and a second displacement vector between each of the control points and the next control point;

[0025] Perform normalization processing on the first displacement vector and the second displacement vector respectively to obtain corresponding first normalization results and second normalization results;

[0026] Determine a displacement vector dot product result corresponding to the first normalization result and the second normalization result, and determine a curvature metric corresponding to the control points based on the displacement vector dot product result;

[0027] Determine an average curvature metric corresponding to each of the control points to obtain a smoothing error value; the numerical value of the smoothing error value is negatively correlated with the magnitude of the color change smoothness of the target style image.

[0028] Optionally, before processing the mapping image, the feature image, and the saliency image based on the feature salient term, the color opponency term, and the feature sensitivity term to obtain a loss function, the method further includes:

[0029] Align the saliency image and the feature image at the pixel level to obtain an alignment result, and then multiply the pixel values in the saliency image by the corresponding pixel values in the feature image to obtain a multiplication result;

[0030] Count the number of pixel points marked as feature regions in the feature image, determine a coincidence degree index based on the multiplication result of each of the pixel values and the number of pixel points, and determine a feature significant item based on the coincidence degree index; the magnitude of the value corresponding to the feature significant item has a negative correlation with the presentation effect of the feature region in the saliency image.

[0031] Optionally, before processing the mapping image, the feature image, and the saliency image based on the feature significant item, the color opposition item, and the feature sensitivity item to obtain a loss function, it further includes:

[0032] Extract the first color values corresponding to the pixel points in each of the feature regions from the mapping image, and determine the corresponding first average modulus length based on each of the first color values;

[0033] Extract the second color values corresponding to each pixel point in each of the non-feature regions from the mapping image, and determine the corresponding second average modulus length based on each of the second color values;

[0034] Determine an absolute difference result based on the first average modulus length and the second average modulus length, and determine a color opposition item based on the absolute difference result; the magnitude of the value of the color opposition item has a negative correlation with the magnitude of the color distinctiveness between the feature region and the non-feature region.

[0035] Optionally, before processing the mapping image, the feature image, and the saliency image based on the feature significant item, the color opposition item, and the feature sensitivity item to obtain a loss function, it further includes:

[0036] Extract each third color value in the feature region from the mapping image, and determine the perceptual distance between every two of the third color values in the color space;

[0037] Set the perceptual distance with the largest value among the perceptual distances as the to-be-processed perceptual distance, and determine a feature sensitivity item based on the to-be-processed perceptual distance; the magnitude of the value of the feature sensitivity item has a negative correlation with the magnitude of the color change span corresponding to the feature region.

[0038] Optionally, using the annealing algorithm and the loss function to adjust the positions of the control points to obtain a position adjustment result, and determining a control point adjustment scheme corresponding to the target style image based on the position adjustment result includes:

[0039] Determine an initial temperature parameter and a temperature reduction rate parameter, and then adjust the initial position direction of the control points using a preset position adjustment direction control rule at the current temperature to obtain a to-be-processed position direction;

[0040] Using the loss function and based on the initial position direction and the position direction to be processed, determine the change amount of the function value corresponding to the loss function value, and determine whether the change amount of the function value is greater than zero. If the change amount of the function value is not greater than zero, set the position direction to be processed as the position adjustment result;

[0041] Determine whether the current temperature is less than a preset temperature threshold. If the current temperature is less than the preset temperature threshold, determine whether the loss function value meets the preset stability condition. If the loss function value meets the preset stability condition, determine a control point adjustment scheme corresponding to the target style image based on the corresponding position adjustment result;

[0042] If the change amount of the function value is greater than zero, determine a new current temperature based on the initial temperature parameter and the temperature reduction rate parameter, and trigger the step of adjusting the initial position direction of the control point by using a preset position adjustment direction control rule at the current temperature.

[0043] In a second aspect, the present application provides an intelligent color mapping device based on feature saliency, including:

[0044] A color mapping table determination module, configured to cluster pixel points corresponding to a target style image to obtain a plurality of control points, determine a color sequence adjustment function corresponding to the control points based on a perceptual uniformity deviation and a smoothing error, and then generate a color mapping table by using the color sequence adjustment function; the perceptual uniformity deviation is the degree of deviation between the current perceptual distance and the preset ideal perceptual distance of each adjacent control point in a perceptually uniform color space; the smoothing error is a color change smoothness error determined by using the angular change of the displacement vectors corresponding to the control points;

[0045] A mapped image acquisition module, configured to perform color mapping on each control point by using the color mapping table to obtain a mapped image, obtain a feature image for identifying a feature region to be highlighted, and process the mapped image by using an MSRMNet network model to obtain a saliency image;

[0046] A loss function determination module, configured to process the mapped image, the feature image, and the saliency image based on a feature saliency term, a color opponency term, and a feature sensitivity term to obtain a loss function; the feature saliency term is the degree of coincidence between the saliency image and the feature image; the color opponency term is the difference between the average color modulus lengths of the feature region and the non-feature region in the feature image; the feature sensitivity term is the maximum distance of the colors in the feature region in the color space;

[0047] A control point adjustment scheme determination module, configured to use an annealing algorithm and the loss function to adjust the positions of the control points, obtain a position adjustment result, and determine a control point adjustment scheme corresponding to the target style image based on the position adjustment result.

[0048] In a third aspect, the present application provides an electronic device, including:

[0049] A memory, configured to store a computer program;

[0050] A processor, configured to execute the computer program to implement the foregoing intelligent color mapping method based on feature saliency.

[0051] In a fourth aspect, the present application provides a computer-readable storage medium, configured to store a computer program, wherein when the computer program is executed by a processor, the foregoing intelligent color mapping method based on feature saliency is implemented.

[0052] As can be seen from the above, before performing intelligent color mapping based on feature saliency, the present application needs to cluster the pixel points corresponding to the target style image to obtain a number of control points, determine a color order adjustment function corresponding to the control points based on the perceptual uniformity deviation and the smoothing error, and then use the color order adjustment function to generate a color mapping table; the perceptual uniformity deviation is the degree of deviation between the perceptual distances of adjacent control points and the ideal uniform state; the smoothing error is the color change smoothness determined by the angular change of the displacement vectors corresponding to the control points; use the color mapping table to perform color mapping on each control point to obtain a mapped image, and obtain a feature image for identifying the feature region to be highlighted, use the MSRMNet network model to process the mapped image to obtain a saliency image; based on the feature saliency term, the color opposition term, and the feature sensitivity term, process the mapped image, the feature image, and the saliency image to obtain a loss function; the feature saliency term is the degree of overlap between the saliency image and the feature image; the color opposition term is the difference between the average color modulus lengths of the feature region and the non-feature region in the feature image; the feature sensitivity term is the maximum distance of the colors in the feature region in the color space; use the annealing algorithm and the loss function to adjust the positions of the control points, obtain a position adjustment result, and determine a control point adjustment scheme corresponding to the target style image based on the position adjustment result.

[0053] It can be seen that in this application, first, the pixel points corresponding to the target style image need to be clustered to obtain several control points, and a color sequence adjustment function corresponding to the control points is determined based on the perceptual uniformity deviation and the smoothing error. Then, a color mapping table is generated using the color sequence adjustment function. Then, the color mapping table is used to perform color mapping on each control point to obtain a mapped image, and a feature image for identifying the feature region to be highlighted is obtained. The MSRMNet network model is used to process the mapped image to obtain a saliency image. Subsequently, based on the feature saliency term, the color opponency term, and the feature sensitivity term, the mapped image, the feature image, and the saliency image are processed to obtain a loss function. Finally, the annealing algorithm and the loss function are used to adjust the positions of the control points to obtain a position adjustment result, and a control point adjustment scheme corresponding to the target style image is determined based on the position adjustment result. In this way, the efficiency of generating the color mapping control point adjustment scheme is improved during the intelligent color mapping based on feature saliency, and thus the efficiency of adjusting the color mapping control points is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] 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 the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the provided drawings without creative efforts.

[0055] Figure 1 It is a flowchart of an intelligent color mapping method based on feature saliency disclosed in this application;

[0056] Figure 2 It is a flowchart of a specific intelligent color mapping method based on feature saliency disclosed in this application;

[0057] Figure 3 It is a schematic diagram of the effects of different style images disclosed in this application; among them, Figure 3 (a) is a schematic diagram of the style image effect corresponding to the first style, Figure 3 (b) is a schematic diagram of the style image effect corresponding to the second style, Figure 3 (c) is a schematic diagram of the style image effect corresponding to the third style;

[0058] Figure 4 It is a schematic diagram of a specific feature image obtained after feature extraction disclosed in this application;

[0059] Figure 5 It is a schematic diagram of the MSRMNet network architecture disclosed in this application;

[0060] Figure 6Schematic diagram of a smart color mapping device based on feature saliency disclosed in this application;

[0061] Figure 7 Structural diagram of an electronic device disclosed in this application. Detailed implementation manners

[0062] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0063] The current data-driven automated design method also has limitations when dealing with complex scientific simulation data. It usually treats all data points equally and applies a unified algorithm for feature expression and definition. However, this approach is difficult to adapt to the non-uniformity and high noise of data distribution, resulting in a lack of robustness in practical applications and the need for a large amount of adjustment or even possible failure in the face of different data types. In addition, existing methods often ignore the need to meet the subjective preferences of different users and lack flexible customization options. Therefore, this application provides a smart color mapping method based on feature saliency, which can improve the efficiency of generating a color mapping control point adjustment scheme during the smart color mapping based on feature saliency, and further improve the efficiency of adjusting the color mapping control points.

[0064] See Figure 1 As shown, the embodiments of the present invention disclose a smart color mapping method based on feature saliency, including:

[0065] Step S11: Cluster the pixel points corresponding to the target style image to obtain a number of control points, and determine a color order adjustment function corresponding to the control points based on the perceptual uniformity deviation and the smoothing error, and then generate a color mapping table using the color order adjustment function; the perceptual uniformity deviation is the degree of deviation between the current perceptual distance and the preset ideal perceptual distance of each adjacent control point in the perceptually uniform color space; the smoothing error is the color change smoothness error determined by using the angular change of the displacement vectors corresponding to each control point.

[0066] In this embodiment, the schematic diagram of performing smart color mapping based on feature saliency is as Figure 2As shown: In the process of intelligent color mapping based on feature saliency, in the embodiments of the present application, first, a target style image needs to be selected according to user preferences and specific application scenarios. Subsequently, the K-Means clustering algorithm is used to cluster the pixel points in the target style image to obtain several control points. It is worth mentioning that in order to meet the requirement of numerical continuity of data, the color distribution of the control points on the color palette must be perceptually uniform. Therefore, in the embodiments of the present application, the perceptually uniform deviation and smooth error proposed in ColorMaker are cited to guide the color order of the control points, and the function expression is as follows:

[0067] ;

[0068] Among them, is the perceptually uniform deviation, is the smooth error.

[0069] It is worth mentioning that if all permutation methods are exhausted, the time complexity is , thus generating a large amount of time cost. Therefore, in the embodiments of the present application, the branch and bound method needs to be used to find the minimum value of . In addition, since the perceived color change generated by a perceptually uniform color palette is proportional to the increase (or decrease) of the data value, that is, the perceptually uniform deviation is essentially the deviation between the current state and the ideal state. Therefore, the specific expression of the perceptually uniform deviation is as follows:

[0070] ;

[0071] Among them, is the standard deviation, is the average value of the perceived distances between all adjacent control points, is the number of control points, is the th control point.

[0072] Specifically, before determining the color order adjustment function corresponding to the control points based on the perceptually uniform deviation and smooth error, and then generating a color mapping table using the color order adjustment function, it may further include: determining control point pairs based on every two control points among the control points, obtaining the current perceived distances between the control point pairs, and then determining the average value of the corresponding distances; determining the distance differences between the perceived distances and the corresponding average values, performing a square operation and summing on the distance differences corresponding to the control point pairs to obtain a summation result, and then determining a calculation result based on the summation result and the number of all control point pairs; determining the perceptually uniform deviation value based on the calculation result and the average value of the distances; the numerical size of the perceptually uniform deviation value is negatively correlated with the size of the perceived uniformity corresponding to the control points.

[0073] Furthermore, since a palette with high curvature produces a steeper color mutation, resulting in visual discontinuity. Therefore, in the embodiments of the present application, it is necessary to measure the angle between displacement vectors at control points to approximate the curvature and evaluate the palette smoothing error. The formula for determining the smoothing error is as follows:

[0074] ;

[0075] where , is the color vector corresponding to the th control point and the th control point in the Lab color space.

[0076] It is worth mentioning that if is 0, it means that the directions of all color vectors are the same, is 1, it means that the directions of all color vectors are opposite (i.e., the angle is 180°). Subsequently, the sorted color list is constructed in a linear mapping manner to obtain a color mapping table , so as to apply images of any style to visualization work, and the effect diagrams of images of different styles are as Figure 3 shown: Figure 3 (a), Figure 3 (b), and Figure 3 (c) all correspond to different styles.

[0077] Specifically, before determining the color order adjustment function corresponding to the control points based on the perceptual uniformity deviation and the smoothing error, and then generating the color mapping table using the color order adjustment function, it may further include: determining the first displacement vector between each control point and the previous control point and the second displacement vector between each control point and the next control point; normalizing the first displacement vector and the second displacement vector respectively to obtain the corresponding first normalization result and the second normalization result; determining the dot product result of the displacement vectors corresponding to the first normalization result and the second normalization result, and determining the curvature metric corresponding to the control points based on the dot product result of the displacement vectors; determining the average value of the corresponding curvature metrics for each control point to obtain the smoothing error value; the numerical size of the smoothing error value is negatively correlated with the smoothness of the color change of the target style image.

[0078] Step S12: Perform color mapping on each of the control points using the color mapping table to obtain a mapped image, and obtain a feature image for identifying a feature region to be highlighted. Process the mapped image using the MSRMNet network model to obtain a saliency image.

[0079] In this embodiment, after obtaining the color mapping table, the embodiments of the present application need to use to complete color mapping of the data, so as to obtain a mapped image . Subsequently, a feature image obtained by the user input through a feature extraction criterion or interaction is acquired , and a schematic diagram of the feature image obtained after feature extraction is as Figure 4 shown. Then, the MSRMNet (Multi-scale skip residual and multi-mixed features network) network model is used to perform saliency detection, so as to output a saliency image . Among them, the MSRMNet network architecture is as Figure 5 shown. It is worth mentioning that the pink images in the legend and the pink images in the encoder and decoder are in a corresponding relationship, both representing the 3×3 convolutional layer. The green images in the legend and the green images in the encoder are in a corresponding relationship, both representing batch normalization; the purple images in the legend and the purple images in the encoder are in a corresponding relationship, both representing the Relu activation function. The blue images in the legend and the blue images in the encoder and decoder are in a corresponding relationship, both representing upsampling.

[0080] Step S13: Based on the feature saliency term, the color opponency term, and the feature sensitivity term, process the mapped image, the feature image, and the saliency image to obtain a loss function; the feature saliency term is the degree of overlap between the saliency image and the feature image; the color opponency term is the difference in the average color modulus length between the feature region and the non-feature region in the feature image; the feature sensitivity term is the maximum distance of the color in the feature region in the color space.

[0081] In this embodiment, after obtaining , and , the embodiments of the present application need to use , and to calculate the loss. It is worth mentioning that during the process of calculating the loss, the embodiments of the present application formalize the entire process into a non-linear constrained optimization problem to balance the three key factors affecting R3: E1: the feature saliency term, E2: the color opponency term, and E3: the feature sensitivity term. Among them, the feature saliency term aims to reduce and the feature region map The differences between them are used to enhance the saliency of the feature region in the mapped image; the color opponency terms are used to avoid the feature region being too similar in color to other regions; the feature sensitivity is used to prevent the numerical changes within the feature region from being ignored. That is, the above three factors are the basis for ensuring that the color palette reveals the feature region and is thus understood by most users.

[0082] In a specific embodiment, it is first necessary to use the minimization of the objective cost function to determine a sequence of positions:

[0083] ;

[0084] In addition, since the control point colors have been fixed, there is only one independent variable, namely the control point position , and the expression of the objective cost function is as follows:

[0085] ;

[0086] Among them, is used to evaluate the saliency degree of the feature region in, is used to describe the color similarity degree between the feature region and the non-feature region in, is used to maintain the consistency of color data and reveal the numerical changes within the feature region.

[0087] Furthermore, the embodiment of the present application needs to adjust the weight of the feature saliency term. That is, in order to obtain a color palette that can truly simulate the human eye's perception of highlighting the feature region, the embodiment of the present application needs to start from the mapped result image to improve the saliency of the feature region [[ID=4,6]]in , and the process is as follows:

[0088] First, analyze the coincidence degree between two single-channel visual importance images [[ID=,53]]and of the same size, and use the coincidence degree as an index to describe the saliency degree of the feature region. Among them, the determination formula of the feature saliency is as follows:

[0089] ;

[0090] ;

[0091] Among them, and are respectively and The width and height, and respectively represent and at the value of , , represents the number of pixel points in the feature region in

[0092] Furthermore, in , in the embodiments of the present application, the product of the values at the same position of two images is accumulated as the degree of consistency between the two images, and then the product of two single-channel images is used to define . In a specific embodiment, when , are both or , which indicates that and do not overlap at all. When , are both , which indicates that and completely overlap.

[0093] Specifically, before obtaining the loss function by processing the mapped image, the feature image, and the saliency image based on the feature saliency term, the color opponency term, and the feature sensitivity term, it may further include: pixel-level alignment of the saliency image and the feature image to obtain an alignment result, and then multiplying the pixel values in the saliency image by the corresponding pixel values in the feature image to obtain a multiplication result; counting the number of pixel points marked as the feature region in the feature image, and determining the coincidence index based on the multiplication result of each pixel value and the number of pixel points, and determining the feature saliency term based on the coincidence index; the numerical size of the feature saliency term has a negative correlation with the presentation effect of the feature region in the saliency image.

[0094] Then, adjust the weight corresponding to the color opponency term, that is, in order to avoid and producing similar colors due to overly similar numerical values, resulting in unclear feature contours, the embodiments of the present application need to add the color opponency term to the penalty function to increase and the distinguishability. In a specific embodiment, the embodiments of the present application define and for two images , and the expression is as follows:<(

[0095] ;

[0096] ;

[0097] Among them, and are respectively and the number of pixel points in is in the three-channel Lab color value at represents the modulus length of the color value . It is worth mentioning that is essentially defined as the absolute value of the difference between the average color modulus length of the feature region and the average color modulus length of the non-feature region. Since is 1 in and 0 in , therefore, only accumulates regions, only accumulates regions.

[0098] Specifically, before obtaining the loss function by processing the mapped image, the feature image, and the saliency image based on the feature saliency term, the color opponency term, and the feature sensitivity term, it may further include: extracting the first color value corresponding to the pixel points in each feature region from the mapped image, and determining the corresponding first average modulus length based on each first color value; extracting the second color value corresponding to each pixel point in each non-feature region from the mapped image, and determining the corresponding second average modulus length based on each second color value; determining the absolute difference result based on the first average modulus length and the second average modulus length, and determining the color opponency term based on the absolute difference result; the numerical value of the color opponency term is negatively correlated with the numerical value of the color distinctiveness between the feature region and the non-feature region.

[0099] Finally, adjust the weight corresponding to the feature sensitivity, that is, if the embodiments of the present application only consider and the difference between them and do not restrict the inside of the region, it may not be able to reveal the numerical changes inside the data. Therefore, the embodiments of the present application add feature sensitivity to make the numerical changes inside the feature have a wider span in color, and the expression of

[0100] ;

[0101] ;

[0102] Among them, is the color set that appears inside the feature region in , and Represents a color value, Represents The perceptual distance in the Lab color space with It is worth mentioning that since the Lab color space is considered to be perceptually uniform, according to the definition of

[0103] As it increases, a larger color interval will be used to represent data changes. Therefore, the data changes within the feature region will also be more sensitive in terms of color, which is more conducive to analyzing the change trend of the feature region data. In this way, the embodiments of the present application can formalize the scalar field color mapping problem into a non-linear optimization problem by using the above mathematical model. Specifically, before obtaining the loss function by processing the mapping image, the feature image, and the saliency image based on the feature significant term, the color opponency term, and the feature sensitivity term, it may further include: extracting each third color value in the feature region from the mapping image, and determining the perceptual distance between every two third color values in the color space; setting the largest perceptual distance among the perceptual distances as the perceptual distance to be processed, and determining the feature sensitivity term based on the perceptual distance to be processed; the numerical size of the feature sensitivity term has a negative correlation with the numerical value of the color change span corresponding to the feature region. As Increases, a larger color interval will be used to represent data changes. Therefore, the data changes within the feature region will also be more sensitive in terms of color, which is more conducive to analyzing the change trend of the feature region data. In this way, the embodiments of the present application can formalize the scalar field color mapping problem into a non-linear optimization problem by using the above mathematical model. Specifically, before obtaining the loss function by processing the mapping image, the feature image, and the saliency image based on the feature significant term, the color opponency term, and the feature sensitivity term, it may further include: extracting each third color value in the feature region from the mapping image, and determining the perceptual distance between every two third color values in the color space; setting the largest perceptual distance among the perceptual distances as the perceptual distance to be processed, and determining the feature sensitivity term based on the perceptual distance to be processed; the numerical size of the feature sensitivity term has a negative correlation with the numerical value of the color change span corresponding to the feature region.

[0104] Step S14: Use the annealing algorithm and the loss function to adjust the positions of the control points, obtain the position adjustment result, and determine the control point adjustment plan corresponding to the target style image based on the position adjustment result.

[0105] In this embodiment, it is necessary to use the annealing algorithm to adjust the positions of the color control points, and make the objective function tend to converge by continuously changing the mapping relationship from the data domain to the color domain. It is worth mentioning that during the process of using the annealing algorithm to adjust the positions of the color control points, the embodiments of the present application first need to start from a relatively high initial "temperature" and then gradually cool down. In each iteration, the algorithm will perturb the current solution by offsetting a small amount to the left or right randomly selected control points, and evaluate the color palette using the cost function. If the cost of the newly obtained solution is lower, the newly obtained solution is adopted. To prevent falling into a local optimum, when the new solution is worse, the following formula is used for probability determination to determine whether to adopt the above solution based on the obtained probability:

[0106] ;

[0107] Wherein, Represents the difference between the current loss and the current optimal state, Represents the current temperature.

[0108] Specifically, the annealing algorithm and the loss function are used to adjust the positions of each control point to obtain a position adjustment result, and a control point adjustment scheme corresponding to the target style image is determined based on the position adjustment result, which may include: determining an initial temperature parameter and a cooling rate parameter, and then adjusting the initial position direction of the control point using a preset position adjustment direction control rule at the current temperature to obtain a position direction to be processed; using the loss function and based on the initial position direction and the position direction to be processed to determine the function value change amount corresponding to the loss function value, and determining whether the function value change amount is greater than zero. If the function value change amount is not greater than zero, the position direction to be processed is set as the position adjustment result; determining whether the current temperature is less than a preset temperature threshold. If the current temperature is less than the preset temperature threshold, determining whether the loss function value meets a preset stability condition. If the loss function value meets the preset stability condition, a control point adjustment scheme corresponding to the target style image is determined based on the corresponding position adjustment result; if the function value change amount is greater than zero, a new current temperature is determined based on the initial temperature parameter and the cooling rate parameter, and the step of adjusting the initial position direction of the control point using a preset position adjustment direction control rule at the current temperature is triggered.

[0109] It can be seen that in the embodiment of the present application, first, the pixel points corresponding to the target style image need to be clustered to obtain several control points, and a color sequence adjustment function corresponding to the control points is determined based on the perceptual uniformity deviation and the smoothing error, and then a color mapping table is generated using the color sequence adjustment function; then, each control point is color-mapped using the color mapping table to obtain a mapped image, and a feature image for identifying a feature region to be highlighted is obtained, and the mapped image is processed using the MSRMNet network model to obtain a saliency image; subsequently, based on the feature saliency term, the color opponency term, and the feature sensitivity term, the mapped image, the feature image, and the saliency image are processed to obtain a loss function; finally, the annealing algorithm and the loss function are used to adjust the positions of each control point to obtain a position adjustment result, and a control point adjustment scheme corresponding to the target style image is determined based on the position adjustment result. In this way, the efficiency of generating the color mapping control point adjustment scheme is improved during the intelligent color mapping based on feature saliency, and further the efficiency of adjusting the color mapping control points is improved.

[0110] Correspondingly, as shown in Figure 6 the present application further provides an intelligent color mapping device based on feature saliency, including:

[0111] A color mapping table determination module 11, which is used to cluster the pixel points corresponding to the target style image to obtain several control points, and determine a color order adjustment function corresponding to the control points based on the perceptual uniformity deviation and the smoothing error, and then generate a color mapping table by using the color order adjustment function; the perceptual uniformity deviation is the degree of deviation between the current perceptual distance of each adjacent control point in the perceptually uniform color space and the preset ideal perceptual distance; the smoothing error is the color change smoothness error determined by using the angular change of the displacement vectors corresponding to the control points;

[0112] A mapped image acquisition module 12, which is used to perform color mapping on each of the control points by using the color mapping table to obtain a mapped image, and acquire a feature image for identifying a feature area to be highlighted, and process the mapped image by using the MSRMNet network model to obtain a saliency image;

[0113] A loss function determination module 13, which is used to process the mapped image, the feature image and the saliency image based on a feature saliency term, a color opposition term and a feature sensitivity term to obtain a loss function; the feature saliency term is the degree of coincidence between the saliency image and the feature image; the color opposition term is the difference between the average color modulus lengths of the feature area and the non-feature area in the feature image; the feature sensitivity term is the maximum distance of the colors in the feature area in the color space;

[0114] A control point adjustment scheme determination module 14, which is used to adjust the positions of the control points by using an annealing algorithm and the loss function to obtain a position adjustment result, and determine a control point adjustment scheme corresponding to the target style image based on the position adjustment result.

[0115] As can be seen from the above, before performing intelligent color mapping based on feature saliency in the embodiments of the present application, it is first necessary to cluster the pixel points corresponding to the target style image to obtain several control points, and determine a color order adjustment function corresponding to the control points based on the perceptual uniformity deviation and the smoothing error. Then, a color mapping table is generated using the color order adjustment function; then, the color mapping table is used to perform color mapping on each control point to obtain a mapped image, and a feature image for identifying the feature region to be highlighted is obtained. The mapped image is processed using the MSRMNet network model to obtain a saliency image; subsequently, based on the feature saliency term, the color opponency term, and the feature sensitivity term, the mapped image, the feature image, and the saliency image are processed to obtain a loss function; finally, the annealing algorithm and the loss function are used to adjust the positions of each control point to obtain a position adjustment result, and a control point adjustment scheme corresponding to the target style image is determined based on the position adjustment result. In this way, the efficiency of generating the color mapping control point adjustment scheme is improved during the process of intelligent color mapping based on feature saliency, and further, the efficiency of adjusting the color mapping control points is improved.

[0116] In some specific embodiments, the intelligent color mapping device based on feature saliency may further include:

[0117] A distance average value determination unit, configured to determine control point pairs based on every two of the control points, obtain the current perceptual distances between the control point pairs, and then determine the distance average values corresponding to the current perceptual distances;

[0118] A calculation result determination unit, configured to determine the distance differences between the perceptual distances and the corresponding distance average values, perform a square operation and summation on the distance differences corresponding to the control point pairs to obtain a summation result, and then determine a calculation result based on the summation result and the number of all the control point pairs;

[0119] A perceptual uniformity deviation value determination unit, configured to determine a perceptual uniformity deviation value based on the calculation result and the distance average value; the numerical size of the perceptual uniformity deviation value is negatively correlated with the perceptual uniformity size corresponding to the control point.

[0120] In some specific embodiments, the intelligent color mapping device based on feature saliency may further include:

[0121] A displacement vector determination unit, configured to determine a first displacement vector between each control point and the previous control point and a second displacement vector between each control point and the next control point;

[0122] A normalization result determination unit, configured to perform normalization processing on the first displacement vector and the second displacement vector respectively, to obtain a corresponding first normalization result and a second normalization result;

[0123] A curvature metric determination unit, configured to determine a dot product result of the displacement vectors corresponding to the first normalization result and the second normalization result, and determine a curvature metric corresponding to the control point based on the dot product result of the displacement vectors;

[0124] A smoothing error value determination unit, configured to determine an average value of the curvature metrics corresponding to the control points based on the curvature metrics of the control points, to obtain a smoothing error value; the numerical value of the smoothing error value is negatively correlated with the smoothness of the color change of the target style image.

[0125] In some specific embodiments, the intelligent color mapping device based on feature saliency may further include:

[0126] A multiplication result determination unit, configured to perform pixel-level alignment on the saliency image and the feature image to obtain an alignment result, and then multiply the pixel values in the saliency image by the corresponding pixel values in the feature image to obtain a multiplication result;

[0127] A feature significant item determination unit, configured to count the number of pixel points marked as feature regions in the feature image, determine a coincidence degree index based on the multiplication results of the pixel values and the number of pixel points, and determine a feature significant item based on the coincidence degree index; the numerical value corresponding to the feature significant item is negatively correlated with the presentation effect of the feature region in the saliency image.

[0128] In some specific embodiments, the intelligent color mapping device based on feature saliency may further include:

[0129] A first modulus length average value determination unit, configured to extract first color values corresponding to pixel points in each of the feature regions from the mapping image, and determine a corresponding first modulus length average value based on the first color values;

[0130] A second modulus length average value determination unit, configured to extract second color values corresponding to each pixel point in each of the non-feature regions from the mapping image, and determine a corresponding second modulus length average value based on the second color values;

[0131] A color opposition item determination unit, configured to determine an absolute difference result based on the first modulus length average value and the second modulus length average value, and determine a color opposition item based on the absolute difference result; the numerical value of the color opposition item is negatively correlated with the numerical value of the color discrimination degree between the feature region and the non-feature region.

[0132] In some specific embodiments, the intelligent color mapping device based on feature saliency may further include:

[0133] A perceived distance determination unit, configured to extract each third color value in the feature region from the mapped image, and determine the perceived distance between every two of the third color values in the color space;

[0134] A feature sensitivity term determination unit, configured to set the largest perceived distance among the perceived distances as the to-be-processed perceived distance, and determine a feature sensitivity term based on the to-be-processed perceived distance; the numerical value of the feature sensitivity term has a negative correlation with the numerical value of the color change span corresponding to the feature region.

[0135] In some specific embodiments, the control point adjustment scheme determination module 14 may specifically include:

[0136] A position and direction adjustment unit, configured to determine an initial temperature parameter and a temperature reduction rate parameter, and then adjust the initial position and direction of the control point using a preset position adjustment direction control rule at the current temperature to obtain a to-be-processed position and direction;

[0137] A position adjustment result determination unit, configured to use the loss function and determine a function value change amount corresponding to the loss function value based on the initial position and direction and the to-be-processed position and direction, and determine whether the function value change amount is greater than zero. If the function value change amount is not greater than zero, set the to-be-processed position and direction as the position adjustment result;

[0138] A control point adjustment scheme subunit, configured to determine whether the current temperature is less than a preset temperature threshold. If the current temperature is less than the preset temperature threshold, determine whether the loss function value satisfies a preset stability condition. If the loss function value satisfies the preset stability condition, determine a control point adjustment scheme corresponding to the target style image based on the corresponding position adjustment result;

[0139] A current temperature determination unit, configured to, if the function value change amount is greater than zero, determine a new current temperature based on the initial temperature parameter and the temperature reduction rate parameter, and trigger the step of adjusting the initial position and direction of the control point using a preset position adjustment direction control rule at the current temperature.

[0140] Furthermore, an embodiment of the present application also discloses an electronic device, Figure 7It is a structural diagram of an electronic device 20 shown according to an exemplary embodiment. The content in the figure should not be regarded as any limitation on the scope of use of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. Among them, the memory 22 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the intelligent color mapping method based on feature saliency disclosed in any of the foregoing embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0141] In this embodiment, the power supply 23 is used to provide operating voltages for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and no specific limitation is imposed on it here; the input / output interface 25 is used to obtain external input data or output data to the outside, and its specific interface type can be selected according to specific application requirements, and no specific limitation is made here.

[0142] In addition, as a carrier for resource storage, the memory 22 can be a read-only memory, a random access memory, a disk, or an optical disc, etc. The resources stored thereon may include an operating system 221, a computer program 222, etc., and the storage method can be short-term storage or permanent storage.

[0143] Among them, the operating system 221 is used to manage and control each hardware device and the computer program 222 on the electronic device 20, and it can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the intelligent color mapping method based on feature saliency executed by the electronic device 20 disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs that can be used to complete other specific tasks.

[0144] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the intelligent color mapping method based on feature saliency disclosed above. For the specific steps of this method, reference can be made to the corresponding content disclosed in the foregoing embodiments, and details are not repeated here.

[0145] In the present specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.

[0146] Those skilled in the art can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been generally described according to their functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0147] The steps of the methods or algorithms described in combination with the embodiments disclosed in this article can be directly implemented by hardware, software modules executed by a processor, or a combination of the two. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0148] Finally, it should also be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the existence of additional identical elements in the process, method, article or device comprising the element.

[0149] The technical solutions provided in this application have been introduced in detail above. Specific examples are used in this article to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. An intelligent color mapping method based on feature saliency, characterized in that Including: Clustering the pixel points corresponding to the target style image to obtain a number of control points, determining a color sequence adjustment function corresponding to the control points based on the perceptual uniform deviation and the smoothing error, and then generating a color mapping table using the color sequence adjustment function; the perceptual uniform deviation is the degree of deviation between the current perceptual distance of each adjacent control point in the perceptually uniform color space and the preset ideal perceptual distance; the smoothing error is the color change smoothness error determined by the angular change of the displacement vectors corresponding to each control point; Performing color mapping on each of the control points using the color mapping table to obtain a mapped image, acquiring a feature image for identifying a feature region to be highlighted, and processing the mapped image using the MSRMNet network model to obtain a saliency image; Processing the mapped image, the feature image, and the saliency image based on a feature saliency term, a color opponency term, and a feature sensitivity term to obtain a loss function; the feature saliency term is the degree of overlap between the saliency image and the feature image; the color opponency term is the difference in the average color modulus length between the feature region and the non-feature region in the feature image; the feature sensitivity term is the maximum distance of the colors in the feature region in the color space; Adjusting the positions of each of the control points using an annealing algorithm and the loss function to obtain a position adjustment result, and determining a control point adjustment scheme corresponding to the target style image based on the position adjustment result.

2. The intelligent color mapping method based on feature saliency according to claim 1, wherein Before determining the color sequence adjustment function corresponding to the control points based on the perceptual uniform deviation and the smoothing error, and then generating a color mapping table using the color sequence adjustment function, it further includes: Determining control point pairs based on every two of the control points, acquiring the current perceptual distance between each control point pair, and then determining the average value of the corresponding distances of each current perceptual distance; Determining the distance difference between each perceptual distance and the corresponding average distance value, performing a square operation and summing the distance differences corresponding to each control point pair to obtain a summation result, and then determining a calculation result based on the summation result and the number of all control point pairs; Determining a perceptual uniform deviation value based on the calculation result and the average distance value; the numerical size of the perceptual uniform deviation value is negatively correlated with the perceptual uniformity size corresponding to the control point.

3. The intelligent color mapping method based on feature saliency according to claim 1, wherein Before determining the color sequence adjustment function corresponding to the control points based on the perceptual uniform deviation and the smoothing error, and then generating a color mapping table using the color sequence adjustment function, it further includes: Determining a first displacement vector between each control point and the previous control point and a second displacement vector between each control point and the next control point; Performing normalization processing on the first displacement vector and the second displacement vector respectively to obtain corresponding first normalization results and second normalization results; Determining the displacement vector dot product result corresponding to the first normalization result and the second normalization result, and determining the curvature measure corresponding to the control point based on the displacement vector dot product result; Determine the average value of the corresponding curvature metrics based on the curvature metrics of each of the control points to obtain a smoothing error value; the numerical magnitude of the smoothing error value is negatively correlated with the smoothness of the color change of the target style image.

4. The intelligent color mapping method based on feature saliency according to claim 1, wherein Before obtaining the loss function by processing the mapped image, the feature image, and the saliency image based on the feature saliency term, the color opposition term, and the feature sensitivity term, it further includes: Perform pixel-level alignment on the saliency image and the feature image to obtain an alignment result, and then multiply the pixel values in the saliency image by the corresponding pixel values in the feature image to obtain a multiplication result; Count the number of pixel points marked as feature regions in the feature image, and determine a coincidence degree index based on the multiplication results of the pixel values and the number of pixel points, and determine a feature saliency term based on the coincidence degree index; the numerical magnitude of the feature saliency term is negatively correlated with the presentation effect of the feature region in the saliency image.

5. The intelligent color mapping method based on feature saliency according to claim 1, wherein Before obtaining the loss function by processing the mapped image, the feature image, and the saliency image based on the feature saliency term, the color opposition term, and the feature sensitivity term, it further includes: Extract the first color values corresponding to the pixel points in each of the feature regions from the mapped image, and determine the corresponding average value of the first modulus length based on the first color values; Extract the second color values corresponding to each pixel point in each of the non-feature regions from the mapped image, and determine the corresponding average value of the second modulus length based on the second color values; Determine an absolute difference result based on the average value of the first modulus length and the average value of the second modulus length, and determine a color opposition term based on the absolute difference result; the numerical magnitude of the color opposition term is negatively correlated with the numerical magnitude of the color distinctness between the feature region and the non-feature region.

6. The intelligent color mapping method based on feature saliency according to claim 1, wherein Before obtaining the loss function by processing the mapped image, the feature image, and the saliency image based on the feature saliency term, the color opposition term, and the feature sensitivity term, it further includes: Extract each third color value in the feature region from the mapped image, and determine the perceptual distance between every two of the third color values in the color space; Set the perceptual distance with the largest numerical value among the perceptual distances as the perceptual distance to be processed, and determine a feature sensitivity term based on the perceptual distance to be processed; the numerical magnitude of the feature sensitivity term is negatively correlated with the numerical magnitude of the color change span corresponding to the feature region.

7. The intelligent color mapping method based on feature saliency according to any one of claims 1 to 6, characterized in that Adjust the positions of the control points by using the annealing algorithm and the loss function to obtain a position adjustment result, and determine a control point adjustment scheme corresponding to the target style image based on the position adjustment result, including: Determine an initial temperature parameter and a temperature reduction rate parameter, and then adjust the initial position direction of the control points by using a preset position adjustment direction control rule at the current temperature to obtain a position direction to be processed; Using the loss function and based on the initial position direction and the position direction to be processed, determine the change amount of the function value corresponding to the loss function value, and determine whether the change amount of the function value is greater than zero. If the change amount of the function value is not greater than zero, set the position direction to be processed as the position adjustment result; Determine whether the current temperature is less than the preset temperature threshold. If the current temperature is less than the preset temperature threshold, determine whether the loss function value meets the preset stability condition. If the loss function value meets the preset stability condition, determine the control point adjustment scheme corresponding to the target style image based on the corresponding position adjustment result; If the change amount of the function value is greater than zero, determine a new current temperature based on the initial temperature parameter and the cooling rate parameter, and trigger the step of adjusting the initial position direction of the control point using the preset position adjustment direction control rule at the current temperature.

8. An intelligent color mapping device based on feature saliency, characterized in that, Comprising: A color mapping table determination module, configured to cluster pixel points corresponding to a target style image to obtain a plurality of control points, and determine a color sequence adjustment function corresponding to the control points based on the perceptual uniformity deviation and the smoothing error, and then generate a color mapping table using the color sequence adjustment function; the perceptual uniformity deviation is the degree of deviation between the current perceptual distance and the preset ideal perceptual distance of each adjacent control point in the perceptually uniform color space; the smoothing error is the color change smoothness error determined by using the angular change of the displacement vector corresponding to each control point; A mapped image acquisition module, configured to perform color mapping on each control point using the color mapping table to obtain a mapped image, and acquire a feature image for identifying a feature region to be highlighted, and process the mapped image using the MSRMNet network model to obtain a saliency image; A loss function determination module, configured to process the mapped image, the feature image, and the saliency image based on a feature saliency term, a color opponency term, and a feature sensitivity term to obtain a loss function; the feature saliency term is the degree of coincidence between the saliency image and the feature image; the color opponency term is the difference between the average color modulus lengths of the feature region and the non-feature region in the feature image; the feature sensitivity term is the maximum distance of the colors in the feature region in the color space; A control point adjustment scheme determination module, configured to adjust the positions of the control points using the annealing algorithm and the loss function to obtain a position adjustment result, and determine a control point adjustment scheme corresponding to the target style image based on the position adjustment result.

9. An electronic device, characterized in that, Comprising: A memory, configured to store a computer program; A processor, configured to execute the computer program to implement the intelligent color mapping method based on feature saliency according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, For storing a computer program, wherein the computer program, when executed by a processor, implements the intelligent color mapping method based on feature saliency according to any one of claims 1 to 7.

Images