Automobile modeling intelligent screening method, device and equipment based on aesthetic entropy

Abstract

The invention provides an automobile modeling intelligent screening method, device and equipment based on aesthetic entropy, and the method comprises the steps: obtaining image data of a plurality of candidate automobile modeling schemes, carrying out the preprocessing of the image data, and extracting three types of modeling features, namely curve features, proportion features and color features; wherein for the curve characteristics representing the modeling fluency, the differential entropy is adopted to calculate the information amount of the continuous curvature change; relative entropy is adopted to calculate the deviation degree between an actual proportion and ideal proportions such as golden section for proportion characteristics representing modeling coordination; and for the color features representing the visual richness of the modeling, calculating the comprehensive information amount of the multi-color channel by adopting the joint entropy. And carrying out weighted fusion on the curve entropy, the proportional entropy and the overall entropy to obtain a comprehensive aesthetic entropy value, and automatically screening out a target modeling scheme meeting aesthetic requirements according to a preset screening condition. And subjective aesthetic judgment is converted into a computable entropy index, so that massive candidate schemes can be quickly processed, and the modeling screening efficiency is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of automotive design, and in particular to an intelligent screening method, device, and equipment for automotive styling based on aesthetic entropy. Background Technology

[0002] With the development of artificial intelligence technology and parametric design tools, automotive styling design is undergoing a digital transformation. Generative design, parametric modeling, and other technologies can generate hundreds or even thousands of candidate styling solutions in a short period of time, posing a serious challenge to traditional methods that rely on manual aesthetic evaluation. Existing methods suffer from significant problems when faced with massive amounts of candidate solutions, including low processing efficiency, high labor costs, and long evaluation cycles, which severely restrict the iteration speed and innovation efficiency of styling design.

[0003] Furthermore, automotive styling aesthetics is a complex, multi-dimensional issue involving factors such as the smoothness and variability of curves, the harmony of overall proportions, and the symmetry and balance of visual elements. From an information theory perspective, excellent styling design needs to achieve a certain balance in visual complexity—overly simple designs appear monotonous and unappealing, while overly complex designs give a chaotic feeling and fail to create a unified aesthetic. However, current technology lacks a unified indicator that can objectively quantify this balance of complexity, making it difficult to achieve precise measurement and automated evaluation of the aesthetic characteristics of styling.

[0004] In view of the above, this application is hereby submitted. Summary of the Invention

[0005] This invention discloses an intelligent screening method, device, and equipment for automobile styling based on aesthetic entropy, aiming to solve the technical problems of existing automobile styling aesthetic evaluation methods being highly subjective, inefficient, and lacking objective quantitative indicators, making it difficult to quickly and automatically screen massive candidate styling schemes.

[0006] The first embodiment of the present invention provides an intelligent screening method for automobile styling based on aesthetic entropy, comprising: Image data of multiple candidate car styling schemes are acquired, and the image data is preprocessed to extract the styling features of each candidate car styling scheme. The styling features include curve features, proportion features and color features. Based on the aforementioned styling features, a multi-dimensional aesthetic entropy value is calculated for each candidate car styling scheme. The multi-dimensional aesthetic entropy value includes curve entropy, proportional entropy, and overall entropy. The curve entropy uses differential entropy to calculate the amount of information about continuous curvature changes. The proportional entropy uses relative entropy to calculate the degree of deviation between the actual proportion and the ideal proportion. The overall entropy uses joint entropy to calculate the comprehensive information of multiple color channels. The comprehensive aesthetic entropy value of each candidate car styling scheme is calculated based on the multi-dimensional aesthetic entropy value. The comprehensive aesthetic entropy value of each candidate car styling scheme is filtered according to preset screening criteria, and the target styling scheme that meets the criteria is output.

[0007] Preferably, the process of extracting the curve features includes: Edge detection is performed on the image data to identify key feature lines in the candidate car styling schemes. The key feature lines include waistline, shoulder line, headlight outline, taillight outline, and side profile. Curve fitting is performed on the key feature lines to obtain the parameterized curve expressions of each key feature line; Uniform sampling is performed along each of the parametric curve expressions, and the curvature value at each sampling point is calculated to generate a curvature value sequence; The curvature value sequence is subjected to a second derivative operation to obtain a second curvature distribution sequence, which serves as the curve feature.

[0008] Preferably, the process of calculating the curve entropy based on the curve characteristics includes: The kernel density of the quadratic curvature distribution sequence is estimated, and the discrete quadratic curvature values ​​are smoothed using a Gaussian kernel function to obtain a continuous probability density function f(κ), where κ represents the quadratic curvature value. The differential entropy value of the probability density function is calculated according to the differential entropy formula H_differential = -∫f(κ)ln f(κ)dκ, where the integration interval is the range of values ​​for the quadratic curvature; The differential entropy value is normalized and mapped to the [0,1] interval to obtain the curve entropy.

[0009] Preferably, the process of extracting the proportional features includes: Identify the main visual element regions in the candidate car styling schemes, including the front area, body area, rear area, window area, and wheel area; Measure the length and width of each of the main visual element regions, calculate the length-to-width ratio between adjacent visual element regions, and generate the actual ratio distribution P. An ideal proportional distribution Q is constructed, which is based on the golden ratio value φ=1.618 and its derived ratios; Extract the overall outline of the candidate car styling scheme, take the longitudinal centerline of the car body as the axis of symmetry, calculate the positional deviation between corresponding points on the outline, and generate the actual symmetry deviation distribution P'. Construct an ideal symmetrical deviation distribution Q', wherein the ideal symmetrical deviation distribution Q' is an ideal distribution in which the deviation values ​​are concentrated at the zero point; The actual proportional distribution P, the ideal proportional distribution Q, the actual symmetry deviation distribution P', and the ideal symmetry deviation distribution Q' are used as the proportional features.

[0010] Preferably, the process of calculating the proportional entropy based on the proportional characteristics includes: The relative entropy between the actual proportional distribution P and the ideal proportional distribution Q is calculated according to the KL divergence formula D_KL(P||Q)=Σp(i)log[p(i) / q(i)] to obtain the proportional deviation, where p(i) is the probability of the i-th proportional value in the actual proportional distribution and q(i) is the corresponding probability in the ideal proportional distribution; The relative entropy between the actual symmetric deviation distribution P' and the ideal symmetric deviation distribution Q' is calculated using the KL divergence formula to obtain the degree of symmetry deviation. The image of the candidate car styling scheme is divided into a preset number of grid regions, and the pixel gray-level weighted centroid position of each grid region is calculated to obtain the visual centroid distribution. Construct an ideal center of gravity distribution, calculate the relative entropy between the visual center of gravity distribution and the ideal center of gravity distribution, and obtain the center of gravity deviation. The proportional entropy is calculated according to the formula E_proportion = 1 / (1 + W1 × D_proportion + W2 × D_symmetry + W3 × D_center of gravity), where D_proportion, D_symmetry, and D_center of gravity are the proportional deviation, symmetry deviation, and center of gravity deviation, respectively, and W1, W2, and W3 are preset sub-weight coefficients.

[0011] Preferably, the process of extracting the color features includes: The image data is converted from the RGB color space to the HSV color space to obtain the hue value H, saturation value S, and lightness value V of each pixel. Histogram statistics are performed on the hue value H, the hue value is divided into a preset number of intervals, the number of pixels in each interval is counted, and the hue edge distribution P(H) is generated. Histogram statistics are performed on the saturation value S and the brightness value V respectively to generate the saturation marginal distribution P(S) and the brightness marginal distribution P(V); Construct a three-dimensional joint histogram of hue, saturation, and brightness, count the number of pixels in each combination interval, and generate a joint distribution P(H,S,V); The hue edge distribution P(H), saturation edge distribution P(S), lightness edge distribution P(V), and combined distribution P(H,S,V) are used as the color features.

[0012] Preferably, the process of calculating the overall entropy includes: The joint entropy value of the joint distribution P(H,S,V) is calculated according to the joint entropy formula H(H,S,V)=-Σp(h,s,v)log2p(h,s,v), where p(h,s,v) is the joint probability of the hue value h, saturation value s, and lightness value v appearing simultaneously. The edge entropy H(H) of the hue edge distribution P(H), the edge entropy H(S) of the saturation edge distribution P(S), and the edge entropy H(V) of the lightness edge distribution P(V) are calculated respectively according to the Shannon entropy formula. The mutual information between color channels is calculated as I(H;S;V) = H(H) + H(S) + H(V) - H(H,S,V), where the mutual information represents the degree of correlation between color channels. Surface normal vectors are extracted from the 3D rendered images of candidate car styling schemes, the directional distribution probability of surface normal vectors is calculated, and the surface reflection entropy is calculated according to the spherical entropy formula. The overall entropy is calculated according to the formula E_overall = α1×H(H,S,V) + α2×I(H;S;V) + α3×E_surface reflection, where α1, α2, and α3 are preset sub-weight coefficients, and the calculation results are normalized.

[0013] Preferably, the process of calculating the comprehensive aesthetic entropy value includes: The curve entropy, proportional entropy and overall entropy are processed by the Min-Max normalization method, and the aesthetic entropy values ​​of each dimension are mapped to the [0,1] interval to obtain the normalized curve entropy E'_curve, the normalized proportional entropy E'_proportion and the normalized overall entropy E'_overall. The comprehensive aesthetic entropy value is calculated according to the formula E_comprehensive = β1×E'_curve + β2×E'_proportion + β3×E'_overall, where β1, β2, and β3 are preset dimension weight coefficients.

[0014] The second embodiment of the present invention provides an intelligent screening device for automobile styling based on aesthetic entropy, comprising: The styling feature extraction unit is used to acquire image data of multiple candidate car styling schemes, preprocess the image data to extract the styling features of each candidate car styling scheme, the styling features including curve features, proportion features and color features; A multi-dimensional aesthetic entropy calculation unit is used to calculate the multi-dimensional aesthetic entropy value of each candidate car styling scheme based on the styling features. The multi-dimensional aesthetic entropy value includes curve entropy, proportional entropy and overall entropy. The curve entropy uses differential entropy to calculate the information content of continuous curvature change. The proportional entropy uses relative entropy to calculate the deviation of the actual proportion from the ideal proportion. The overall entropy uses joint entropy to calculate the comprehensive information content of multiple color channels. The comprehensive aesthetic entropy value calculation unit is used to calculate the comprehensive aesthetic entropy value of each candidate car styling scheme based on the multi-dimensional aesthetic entropy values. The target styling scheme selection unit is used to filter the comprehensive aesthetic entropy value of each candidate car styling scheme according to preset selection conditions, and output the target styling scheme that meets the conditions.

[0015] The third embodiment of the present invention provides an intelligent screening device for automobile styling based on aesthetic entropy, characterized in that it includes a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement an intelligent screening method for automobile styling based on aesthetic entropy as described in any of the above embodiments.

[0016] This invention provides an intelligent automotive styling selection method, apparatus, and device based on aesthetic entropy. It acquires image data of multiple candidate automotive styling schemes and preprocesses them to extract three types of styling features: curve features, proportion features, and color features. Specifically, for curve features representing the smoothness of the styling, differential entropy is used to calculate the information content of continuous curvature changes; for proportion features representing the harmony of the styling, relative entropy is used to calculate the deviation of the actual proportion from ideal proportions such as the golden ratio; and for color features representing the visual richness of the styling, joint entropy is used to calculate the comprehensive information content of multiple color channels. The curve entropy, proportion entropy, and overall entropy are weighted and fused to obtain a comprehensive aesthetic entropy value, and target styling schemes that meet aesthetic requirements are automatically selected according to preset selection conditions. This transforms subjective aesthetic judgment into a calculable entropy value index, enabling rapid processing of massive candidate schemes and significantly improving the efficiency of styling selection. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an intelligent screening method for automobile styling based on aesthetic entropy provided in the first embodiment of the present invention. Figure 2 This is a schematic diagram of a module of an intelligent automotive styling screening device based on aesthetic entropy provided in the second embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] To better understand the technical solution of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] This invention discloses an intelligent screening method, device, and equipment for automobile styling based on aesthetic entropy, aiming to solve the technical problems of existing automobile styling aesthetic evaluation methods being highly subjective, inefficient, and lacking objective quantitative indicators, making it difficult to quickly and automatically screen massive candidate styling schemes.

[0021] Please see Figure 1The first embodiment of the present invention provides an intelligent screening method for automobile styling based on aesthetic entropy, which can be executed by an intelligent screening device for automobile styling based on aesthetic entropy (hereinafter referred to as the screening device or system), specifically, by one or more processors within the screening device, to at least implement the following steps: S101, acquire image data of multiple candidate car styling schemes, preprocess the image data to extract the styling features of each candidate car styling scheme, the styling features including curve features, proportion features and color features; In this embodiment, the screening device or system can be a terminal with data processing capabilities, such as a server, workstation, or desktop computer. The screening device can be equipped with a corresponding operating system and application software, and the functions required in this embodiment can be realized through the combination of the operating system and application software.

[0022] In this embodiment, image data of multiple candidate vehicle styling schemes are first acquired. This image data can originate from styling schemes automatically generated by parametric design tools, styling schemes output by generative artificial intelligence models, or styling schemes manually drawn by designers using computer-aided design software. The specific forms of the image data include two-dimensional views such as the front view, side view, rear view, top view, and 45-degree perspective view of the vehicle styling, or three-dimensional rendered images generated under set lighting conditions. After acquiring the image data, it is preprocessed to extract the styling features of each candidate vehicle styling scheme. These styling features include three dimensions: curve features, proportion features, and color features.

[0023] When extracting curve features, the image data is first processed for edge detection. The Canny edge detection algorithm or other mature edge extraction methods can be used to identify key feature lines of candidate car styling schemes from the image. Key feature lines are contour lines in car styling that have a significant impact on the overall visual effect. These include the waistline running along the side of the car body, the shoulder line on the upper part of the body, the contour lines of the headlights and taillights, and the side profile lines extending from the front to the rear. After identifying the key feature lines, parametric curve models such as B-spline curves or Bézier curves are used to fit each key feature line to obtain parametric curve expressions that accurately describe the shape of each feature line. Then, uniform sampling is performed along each parametric curve expression. At each sampling point, the curvature value is calculated according to the parametric equation of the curve. The curvature values ​​of all sampling points are arranged sequentially to generate a curvature value sequence. The curvature value reflects the degree of curvature of the curve at each point; a larger curvature value indicates a more severe curvature. To further capture the dynamic characteristics of curve changes, a second derivative operation is performed on the curvature value sequence to obtain a second curvature distribution sequence as the final curve feature. The second curvature can reflect the changing trend of curvature itself, thus more precisely representing the balance between the smoothness and richness of the shape curve.

[0024] When extracting proportional features, the candidate car styling images are first processed by semantic segmentation or region recognition, dividing the images into several main visual element regions, including the front, body, rear, window, and wheel regions. For each identified main visual element region, its length and width are measured, and the length-to-width ratio between adjacent visual element regions is calculated, such as the ratio of the front length to the body length, and the ratio of the window height to the body height. All ratio values ​​are then aggregated to generate the actual proportion distribution P. Simultaneously, an ideal proportion distribution Q is constructed as an evaluation benchmark. The ideal proportion distribution is based on the golden ratio φ≈1.618 and its derivatives, as the golden ratio is widely considered the most visually aesthetically pleasing proportion. Furthermore, the overall outline of the candidate car styling scheme is extracted. Using the longitudinal centerline of the body as the axis of symmetry, the positional deviation values ​​between corresponding points on the left and right sides of the outline are calculated, and all deviation values ​​are aggregated to generate the actual symmetry deviation distribution P'. Ideally, the car styling should have good left-right symmetry; therefore, an ideal symmetry deviation distribution Q' is constructed. This distribution is an ideal distribution where the deviation values ​​are concentrated near the zero point, representing a perfect symmetry state. Finally, the actual proportional distribution P, the ideal proportional distribution Q, the actual symmetric deviation distribution P', and the ideal symmetric deviation distribution Q' are used together as proportional features for subsequent calculation of proportional entropy.

[0025] When extracting color features, the image data is first converted from the RGB color space to the HSV color space. The HSV color space decomposes color information into three independent channels: hue, saturation, and lightness. Compared to the RGB color space, it better aligns with human color perception and facilitates the analysis of color distribution characteristics. After conversion, the hue value H, saturation value S, and lightness value V of each pixel in the image are obtained. Histogram statistics are performed on the hue value H, dividing the range of hue values ​​into a preset number of intervals. The number of pixels falling into each interval is counted and normalized to generate the hue edge distribution P(H). The same method is used to perform histogram statistics on the saturation value S and the lightness value V, respectively, generating the saturation edge distribution P(S) and the lightness edge distribution P(V). The edge distribution reflects the probability distribution of a single color channel. To capture the correlation between different color channels, a three-dimensional joint histogram of hue, saturation, and brightness is further constructed. The value ranges of the three channels are divided into several intervals to form a three-dimensional grid. The number of pixels falling into each combination interval is counted and normalized to generate a joint distribution P(H,S,V). Finally, the hue edge distribution P(H), saturation edge distribution P(S), brightness edge distribution P(V), and joint distribution P(H,S,V) are used together as color features for subsequent calculation of overall entropy.

[0026] S102, calculate the multi-dimensional aesthetic entropy value of each candidate car styling scheme based on the styling features. The multi-dimensional aesthetic entropy value includes curve entropy, proportional entropy and overall entropy. The curve entropy uses differential entropy to calculate the information content of continuous curvature change. The proportional entropy uses relative entropy to calculate the deviation of the actual proportion from the ideal proportion. The overall entropy uses joint entropy to calculate the comprehensive information content of multiple color channels. It should be noted that after extracting the styling features, the multi-dimensional aesthetic entropy value of each candidate car styling scheme is calculated based on the extracted curve features, proportion features, and color features. The multi-dimensional aesthetic entropy value includes three dimensions: curve entropy, proportion entropy, and overall entropy. These dimensions quantify the aesthetic characteristics of the car styling from three aspects: the smoothness and variability of the styling lines, the harmony and symmetry of the styling proportions, and the richness and unity of the styling colors. To address the different mathematical properties of the three types of styling features, this embodiment employs differentiated entropy calculation methods: curve entropy uses differential entropy to calculate the information content of continuous curvature changes; proportion entropy uses relative entropy to calculate the deviation between the actual proportion and the ideal proportion; and overall entropy uses joint entropy to calculate the comprehensive information content of multiple color channels.

[0027] When calculating curve entropy, since the quadratic curvature distribution sequence consists of discretely sampled data points, kernel density estimation is first performed on the sequence to more accurately estimate its continuous probability distribution characteristics. Kernel density estimation is a non-parametric probability density estimation method that can infer the overall probability density distribution shape from finite sample data. Specifically, a Gaussian kernel function is used to smooth the discrete quadratic curvature values. The Gaussian kernel function has good smoothing properties and excellent mathematical properties, effectively suppressing the influence of sampling noise. The continuous probability density function f(κ) describing the quadratic curvature distribution is obtained through kernel density estimation, where κ represents the quadratic curvature value. After obtaining the continuous probability density function, the differential entropy value is calculated using the differential entropy formula H_differential = -∫f(κ)lnf(κ)dκ, where the integration interval is the actual range of quadratic curvature values. Differential entropy is a generalization of Shannon entropy to continuous random variables and can measure the amount of information or uncertainty contained in a continuous probability distribution. For automotive styling curves, a higher differential entropy value indicates a richer and more diverse curvature variation, while a lower value indicates a simpler and smoother curvature variation. Since the original differential entropy values ​​of different candidate schemes may fall within different numerical ranges, the differential entropy values ​​are normalized to facilitate subsequent comparison and fusion calculations. The Min-Max normalization method is used to map them to the [0,1] interval, ultimately obtaining the curve entropy.

[0028] The core idea in calculating proportional entropy is to use relative entropy to measure the degree of difference between the actual proportions of a design and the ideal aesthetic proportions. Relative entropy, also known as KL divergence, is a classic method in information theory for measuring the difference between two probability distributions; a larger KL divergence value indicates a more significant difference between the two distributions. First, the relative entropy between the actual proportional distribution P and the ideal proportional distribution Q is calculated using the KL divergence formula D_KL(P||Q)=Σp(i)log[p(i) / q(i)], yielding the proportional deviation. Here, p(i) is the probability of the i-th proportional value appearing in the actual proportional distribution, and q(i) is the probability of the corresponding proportional value in the ideal proportional distribution. The proportional deviation reflects the degree to which the actual proportional relationship of each visual element in the design matches ideal proportions such as the golden ratio; a smaller deviation indicates that the proportions of the design are closer to the ideal aesthetic standard. Next, using the same KL divergence calculation method, the relative entropy between the actual symmetry deviation distribution P' and the ideal symmetry deviation distribution Q' is calculated, yielding the symmetry deviation. The symmetry deviation reflects the degree of left-right symmetry of the design outline; a smaller deviation indicates better symmetry. Furthermore, the images of candidate car styling schemes are divided into a predetermined number of grid regions, such as 8×8 or 16×16 grids. The weighted centroid position of the pixel grayscale values ​​within each grid region is calculated, and the centroid positions of all grids are summed to obtain the visual center of gravity distribution. Simultaneously, an ideal center of gravity distribution is constructed. This ideal distribution is typically set as a uniform distribution of the center of gravity or a distribution conforming to specific aesthetic principles. The relative entropy between the visual center of gravity distribution and the ideal center of gravity distribution is calculated to obtain the center of gravity deviation. The center of gravity deviation reflects the degree of balance in the distribution of the visual weight of the styling. Finally, the proportional entropy is calculated using the formula E_proportion = 1 / (1 + W1 × D_proportion + W2 × D_symmetry + W3 × D_center of gravity), where D_proportion, D_symmetry, and D_center of gravity are the proportional deviation, symmetry deviation, and center of gravity deviation, respectively, and W1, W2, and W3 are predetermined sub-weight coefficients used to adjust the relative importance of the three deviations in the proportional entropy calculation. The formula uses a reciprocal form to convert the deviation into an entropy value, so that the smaller the deviation, the higher the entropy of the corresponding proportion, which conforms to the logical relationship that "the closer to the ideal proportion, the higher the aesthetic value".

[0029] When calculating the overall entropy, the richness of color distribution, the coordination between color channels, and the light and shadow characteristics of the surface are comprehensively considered. First, the joint entropy value of the joint distribution P(H,S,V) is calculated according to the joint entropy formula H(H,S,V)=-Σp(h,s,v)log2p(h,s,v), where the summation iterates through all combinations of hue value h, saturation value s, and lightness value v, and p(h,s,v) is the joint probability of the corresponding combination occurring simultaneously. Joint entropy can measure the overall information content of the three color channels; the higher the joint entropy value, the richer and more diverse the color combinations. Then, the edge entropy H(H) of the hue edge distribution P(H), the edge entropy H(S) of the saturation edge distribution P(S), and the edge entropy H(V) of the lightness edge distribution P(V) are calculated respectively according to the Shannon entropy formula H=-Σp(x)log2p(x). Edge entropy reflects the uniformity of the distribution of a single color channel. Based on this, the mutual information between color channels is calculated as I(H;S;V) = H(H) + H(S) + H(V) - H(H,S,V). Mutual information characterizes the correlation between the hue, saturation, and brightness channels. A higher mutual information value indicates a closer correlation between color channels, suggesting better inherent harmony in the color scheme; a lower mutual information value indicates relative independence between channels, potentially leading to a lack of overall unity in the color combination. Furthermore, surface normal vectors are extracted from the 3D rendered images of candidate car styling schemes. Normal vectors describe the orientation of points on the styling surface. The probability distribution of surface normal vectors in 3D space is statistically analyzed, and surface reflection entropy is calculated using the spherical entropy formula. Spherical entropy is an entropy measure defined on a sphere, suitable for describing the information content of directional distribution. Surface reflection entropy reflects the complexity of the styling surface; a higher surface reflection entropy indicates richer variations in the orientation of the styling surface, resulting in more diverse lighting effects. Finally, the overall entropy is calculated using the formula E_overall = α1 × H(H,S,V) + α2 × I(H;S;V) + α3 × E_surface reflection, where α1, α2, and α3 are preset sub-weighting coefficients used to adjust the contribution ratios of joint entropy, mutual information, and surface reflection entropy to the overall entropy. After calculation, the overall entropy is normalized and mapped to the [0,1] interval to facilitate subsequent calculation of the comprehensive aesthetic entropy value.

[0030] S103, Calculate the comprehensive aesthetic entropy value of each candidate car styling scheme based on the multi-dimensional aesthetic entropy value; It should be noted that after calculating the curve entropy, proportional entropy, and overall entropy, the aesthetic entropy values ​​of the three dimensions need to be fused to obtain a comprehensive aesthetic entropy value that can reflect the overall aesthetic characteristics of the car's styling. Since curve entropy, proportional entropy, and overall entropy are calculated using three different methods—differential entropy, relative entropy, and joint entropy—the dimensions and ranges of their original values ​​may differ. Directly weighting and fusing these values ​​could lead to certain dimensions dominating the fused result due to their larger numerical ranges, affecting the rationality and accuracy of the comprehensive evaluation. Therefore, before performing the fusion calculation, the aesthetic entropy values ​​of the three dimensions are first normalized to eliminate the influence of dimensional differences.

[0031] Specifically, the curve entropy, proportional entropy, and overall entropy are processed using the Min-Max normalization method. Min-Max normalization is a linear transformation method whose basic principle is to scale the original data proportionally to a specified target interval. For the aesthetic entropy value of any dimension, the entropy values ​​of all candidate car styling schemes in that dimension are first calculated to determine the maximum and minimum values. Then, the entropy values ​​of each candidate scheme are mapped to the [0,1] interval according to the formula E'=(E-E_min) / (E_max-E_min), where E is the original entropy value, E_min and E_max are the minimum and maximum values ​​of the entropy values ​​of all candidate schemes in that dimension, and E' is the normalized entropy value. Through the above processing, the normalized curve entropy E'_curve, the normalized proportional entropy E'_proportion, and the normalized overall entropy E'_overall are obtained respectively. All three normalized entropy values ​​are within the [0,1] interval, which facilitates subsequent weighted fusion calculations.

[0032] After normalization, the comprehensive aesthetic entropy value of each candidate car styling scheme is calculated according to the formula E_comprehensive = β1 × E'_curve + β2 × E'_proportion + β3 × E'_overall, where β1, β2, and β3 are preset dimension weight coefficients, representing the relative importance of curve entropy, proportion entropy, and overall entropy in the comprehensive aesthetic evaluation, respectively, and satisfying the constraint β1 + β2 + β3 = 1. The dimension weight coefficients can be adjusted according to specific application scenarios and evaluation objectives. For example, for sporty cars emphasizing dynamic and smooth styling, the weight of curve entropy β1 can be appropriately increased; for business cars emphasizing stable and harmonious styling, the weight of proportion entropy β2 can be appropriately increased; and for personalized cars emphasizing visual impact and color expression, the weight of overall entropy β3 can be appropriately increased. In the default configuration, all three weight coefficients can be set to 1 / 3, indicating that the three dimensions have equal importance in the comprehensive evaluation. Furthermore, this invention also allows users to customize the dimension weight coefficients according to their personal aesthetic preferences through a human-computer interaction interface. The adjustment range is [0,1], and the adjusted value must satisfy the constraint that the sum of the three weight coefficients equals 1. This allows the comprehensive aesthetic entropy value to better reflect the aesthetic preferences of a specific user group or market segment. Through the above weighted fusion calculation, the comprehensive aesthetic entropy value of each candidate car styling scheme is finally obtained. This value is also within the [0,1] range. The closer the value is to the preset ideal entropy value range, the higher the comprehensive aesthetic value of the styling.

[0033] S104: Based on preset screening conditions, the comprehensive aesthetic entropy value of each candidate car styling scheme is screened, and the target styling scheme that meets the conditions is output.

[0034] It should be noted that after calculating the comprehensive aesthetic entropy value of each candidate car styling scheme, the comprehensive aesthetic entropy value is filtered according to preset screening conditions. The system automatically identifies and outputs the target styling scheme that meets aesthetic requirements from among numerous candidate schemes. The screening conditions are based on the correspondence between aesthetic entropy and human aesthetic perception. According to the principles of information theory aesthetics, excessively low entropy values ​​correspond to overly simple and monotonous styling, lacking visual appeal; excessively high entropy values ​​correspond to overly complex and chaotic styling, making it difficult to achieve a unified and harmonious aesthetic; while entropy values ​​at a moderate level correspond to "orderly and rich" styling features, achieving a good balance between simplicity and complexity, and best meeting human aesthetic preferences. This invention provides multiple flexible screening methods to adapt to different application scenarios and evaluation needs.

[0035] The first screening method is the target interval screening method. This method sets a target interval [E_min, E_max] for the comprehensive aesthetic entropy value, and selects all candidate car styling schemes whose comprehensive aesthetic entropy values ​​fall within this interval as the target styling schemes. The upper and lower bounds of the target interval can be customized according to specific design standards and market positioning. By default, the target interval is set to [0.4, 0.6], which corresponds to a moderate level of aesthetic entropy. Styling schemes falling within this interval are neither too simple nor too complex in terms of visual complexity, possessing high aesthetic value. The advantage of this screening method is that it clearly defines the boundaries of aesthetic acceptance, making it suitable for scenarios requiring batch screening of all schemes that meet basic aesthetic requirements. The number of screening results depends on the actual distribution of candidate schemes.

[0036] The second screening method is the target entropy approximation method. This method sets a specific target entropy value, E_target, as a reference point for the ideal aesthetic standard. The absolute value of the difference between the comprehensive aesthetic entropy value of each candidate car styling scheme and this target entropy value is calculated. The smaller the absolute value of the difference, the closer the aesthetic characteristics of the scheme are to the ideal standard. All candidate schemes are sorted from smallest to largest based on the absolute value of the difference, and a preset number N of schemes are selected as the target styling schemes. The default value of the target entropy value E_target can be set to 0.5, indicating that the ideal styling should be at the middle level of aesthetic entropy. The preset number N can be set according to actual needs; for example, setting it to 10 means selecting the top 10 styling schemes that are closest to the ideal aesthetic standard. The advantage of this screening method is that it can precisely control the number of output schemes, and the screening results are sorted according to their closeness to the ideal standard, making it convenient for designers to review by priority. It is suitable for scenarios that require quickly locating the optimal scheme from a large number of candidate schemes.

[0037] The third screening method is the proportional ranking screening method. First, all candidate car styling schemes are ranked according to their comprehensive aesthetic entropy value. The ranking direction can be chosen from high to low or low to high, depending on the needs. Then, the schemes with the highest ranking (P%) are selected as the target styling schemes. For example, setting P=20 means selecting the top 20% of styling schemes based on their comprehensive aesthetic entropy value. The advantage of this screening method is that the number of selected results is proportional to the total number of candidate schemes, allowing it to adapt to different sizes of candidate scheme sets. It is suitable for scenarios where the candidate range needs to be reduced by a fixed proportion in the initial screening stage. The choice of ranking direction depends on the specific aesthetic evaluation criteria. If a higher entropy value is considered to represent greater visual richness, then ranking from high to low is chosen. If an entropy value closer to a specific range is considered to represent higher aesthetic value, then it can be combined with the first or second screening method.

[0038] In practical applications, the three screening methods described above can be used individually or in combination to achieve multi-level progressive screening. For example, the target interval screening method can be used first to eliminate schemes with significantly high or low comprehensive aesthetic entropy values. Then, the target entropy value approximation method can be used to further screen the remaining schemes to select the schemes that are closest to the ideal standard for the designer's final review. After screening, the system outputs a list of target design schemes that meet the screening criteria. It can also output auxiliary information such as the comprehensive aesthetic entropy value, aesthetic entropy values ​​of each dimension, and ranking of each target scheme, providing comprehensive data support for design decisions.

[0039] Please see Figure 2 The second embodiment of the present invention provides an intelligent screening device for automobile styling based on aesthetic entropy, comprising: The styling feature extraction unit is used to acquire image data of multiple candidate car styling schemes, preprocess the image data to extract the styling features of each candidate car styling scheme, the styling features including curve features, proportion features and color features; A multi-dimensional aesthetic entropy calculation unit is used to calculate the multi-dimensional aesthetic entropy value of each candidate car styling scheme based on the styling features. The multi-dimensional aesthetic entropy value includes curve entropy, proportional entropy and overall entropy. The curve entropy uses differential entropy to calculate the information content of continuous curvature change. The proportional entropy uses relative entropy to calculate the deviation of the actual proportion from the ideal proportion. The overall entropy uses joint entropy to calculate the comprehensive information content of multiple color channels. The comprehensive aesthetic entropy value calculation unit is used to calculate the comprehensive aesthetic entropy value of each candidate car styling scheme based on the multi-dimensional aesthetic entropy values. The target styling scheme selection unit is used to filter the comprehensive aesthetic entropy value of each candidate car styling scheme according to preset selection conditions, and output the target styling scheme that meets the conditions.

[0040] The third embodiment of the present invention provides an intelligent screening device for automobile styling based on aesthetic entropy, characterized in that it includes a memory and a processor, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement an intelligent screening method for automobile styling based on aesthetic entropy as described in any of the above embodiments.

[0041] This invention provides an intelligent automotive styling selection method, apparatus, and device based on aesthetic entropy. It acquires image data of multiple candidate automotive styling schemes and preprocesses them to extract three types of styling features: curve features, proportion features, and color features. Specifically, for curve features representing the smoothness of the styling, differential entropy is used to calculate the information content of continuous curvature changes; for proportion features representing the harmony of the styling, relative entropy is used to calculate the deviation of the actual proportion from ideal proportions such as the golden ratio; and for color features representing the visual richness of the styling, joint entropy is used to calculate the comprehensive information content of multiple color channels. The curve entropy, proportion entropy, and overall entropy are weighted and fused to obtain a comprehensive aesthetic entropy value, and target styling schemes that meet aesthetic requirements are automatically selected according to preset selection conditions. This transforms subjective aesthetic judgment into a calculable entropy value index, enabling rapid processing of massive candidate schemes and significantly improving the efficiency of styling selection.

[0042] Exemplary examples show that the computer program described in the third and fourth embodiments of the present invention can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in implementing an intelligent screening device for car styling based on aesthetic entropy. For example, the apparatus described in the second embodiment of the present invention.

[0043] The processor referred to can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the aforementioned intelligent screening method for automotive styling based on aesthetic entropy, connecting various parts of the entire system through various interfaces and circuits to implement the intelligent screening method for automotive styling based on aesthetic entropy.

[0044] The memory can be used to store the computer program and / or modules. The processor, by running or executing the computer program and / or modules stored in the memory, and by calling the data stored in the memory, implements various functions of an intelligent automotive styling selection method based on aesthetic entropy. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, text conversion function, etc.), etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, text message data, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0045] If the implemented module is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0046] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0047] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An aesthetic entropy-based intelligent screening method for automobile styling, characterized in that, The method comprises the following steps: acquiring image data of a plurality of candidate automobile styling schemes, and pre-processing the image data to extract styling features of each candidate automobile styling scheme, the styling features including curve features, proportion features, and color features; calculating multi-dimensional aesthetic entropy values of each candidate automobile styling scheme according to the styling features, the multi-dimensional aesthetic entropy values including curve entropy, proportion entropy, and overall entropy, wherein the curve entropy calculates the information amount of continuous curvature change by using differential entropy, the proportion entropy calculates the deviation degree of actual proportions from ideal proportions by using relative entropy, and the overall entropy calculates the comprehensive information amount of multiple color channels by using joint entropy; calculating comprehensive aesthetic entropy values of each candidate automobile styling scheme according to the multi-dimensional aesthetic entropy values; screening the comprehensive aesthetic entropy values of each candidate automobile styling scheme according to a preset screening condition, and outputting a target styling scheme meeting the condition.

2. The method of claim 1, wherein the method is characterized by, The process of extracting the curve features comprises the following steps: performing edge detection on the image data to identify key feature lines in the candidate automobile styling scheme, the key feature lines including waist lines, shoulder lines, front lamp contour lines, rear lamp contour lines, and side body lines; performing curve fitting on the key feature lines to obtain parameterized curve expressions of the key feature lines; uniformly sampling along each parameterized curve expression to calculate curvature values at each sampling point, and generating a curvature value sequence; performing a second-order derivative operation on the curvature value sequence to obtain a second-order curvature distribution sequence as the curve features.

3. The method of claim 2, wherein the method further comprises: The process of calculating the curve entropy according to the curve features comprises the following steps: performing kernel density estimation on the second-order curvature distribution sequence, and performing smoothing processing on discrete second-order curvature values by using a Gaussian kernel function to obtain a continuous probability density function f (κ), wherein κ represents a second-order curvature value; calculating a differential entropy value of the probability density function according to a differential entropy formula H_diff=-∫f(κ)ln f(κ)dκ, wherein the integral interval is the value range of the second-order curvature; performing normalization processing on the differential entropy value to map it to the interval [0, 1] to obtain the curve entropy.

4. The method of claim 1, wherein the method is characterized by, The process of extracting the proportion features comprises the following steps: identifying main visual element regions in the candidate automobile styling scheme, the main visual element regions including a vehicle head region, a vehicle body region, a vehicle tail region, a vehicle window region, and a vehicle wheel region; measuring the length and width dimensions of each main visual element region, calculating the length-width proportion values between adjacent visual element regions, and generating an actual proportion distribution P; constructing an ideal proportion distribution Q, the ideal proportion distribution Q being constructed based on a golden section proportion value φ=1.618 and derived proportions thereof; extracting the overall contour line of the candidate automobile styling scheme, taking the vehicle body longitudinal mid-axis as the symmetry axis, calculating the position deviation values between corresponding points on the contour line, and generating an actual symmetry deviation distribution P'; constructing an ideal symmetry deviation distribution Q', the ideal symmetry deviation distribution Q' being an ideal distribution with deviation values centered on zero; taking the actual proportion distribution P, the ideal proportion distribution Q, the actual symmetry deviation distribution P', and the ideal symmetry deviation distribution Q' as the proportion features.

5. The method of claim 4, wherein the aesthetic entropy is calculated by the following equation: ###0001### wherein, A is the aesthetic entropy, S is the entropy of the car model, and P is the probability of the car model. The process of calculating the proportion entropy according to the proportion features comprises the following steps: According to a KL divergence formula D_KL(P||Q)=Σp(i)log[p(i) / q(i)], a relative entropy between the actual proportion distribution P and an ideal proportion distribution Q is calculated to obtain a proportion deviation degree, where p(i) is a probability of an i-th proportion value in the actual proportion distribution, and q(i) is a corresponding probability in the ideal proportion distribution; According to a KL divergence formula, a relative entropy between the actual symmetric deviation distribution P' and an ideal symmetric deviation distribution Q' is calculated to obtain a symmetric deviation degree; An image of the candidate automobile modeling scheme is divided into a preset number of grid regions, a pixel gray weighted centroid position of each grid region is calculated, and a visual barycenter distribution is obtained; An ideal barycenter distribution is constructed, a relative entropy between the visual barycenter distribution and the ideal barycenter distribution is calculated, and a barycenter deviation degree is obtained; According to a formula E_ proportion=1 / (1+W1×D_ proportion+W2×D_ symmetric+W3×D_ barycenter), where D_ proportion, D_ symmetric and D_ barycenter are the proportion deviation degree, the symmetric deviation degree and the barycenter deviation degree respectively, and W1, W2 and W3 are preset sub-weight coefficients.

6. The method of claim 1, wherein the method is characterized by: The process of extracting the color feature includes: The image data is converted from an RGB color space to an HSV color space to obtain hue H, saturation S and brightness V of each pixel; The hue H is subjected to histogram statistics, the hue is divided into a preset number of intervals, the number of pixels in each interval is counted, and a hue edge distribution P(H) is generated; The saturation S and the brightness V are subjected to histogram statistics respectively, and a saturation edge distribution P(S) and a brightness edge distribution P(V) are generated; A three-dimensional joint histogram of the hue, the saturation and the brightness is constructed, the number of pixels in each combined interval is counted, and a joint distribution P(H, S, V) is generated; The hue edge distribution P(H), the saturation edge distribution P(S), the brightness edge distribution P(V) and the joint distribution P(H, S, V) are taken as the color feature.

7. The method of claim 6, wherein the method further comprises: The process of calculating the overall entropy includes: According to a joint entropy formula H(H, S, V)=-Σp(h, s, v)log2p(h, s, v), a joint entropy value of the joint distribution P(H, S, V) is calculated, where p(h, s, v) is a joint probability of simultaneous occurrence of the hue h, the saturation s and the brightness v; According to a Shannon entropy formula, an edge entropy H(H) of the hue edge distribution P(H), an edge entropy H(S) of the saturation edge distribution P(S) and an edge entropy H(V) of the brightness edge distribution P(V) are respectively calculated; Mutual information I(H; S; V)=H(H)+H(S)+H(V)-H(H, S, V) between color channels is calculated, and the mutual information represents a degree of correlation between the color channels; A surface normal vector of a three-dimensional rendering image of the candidate automobile modeling scheme is extracted, a directional distribution probability of the surface normal vector is calculated, and a surface reflection entropy is calculated according to a spherical entropy formula; The overall entropy is calculated according to a formula E_total = α1*H(H, S, V) + α2*I(H; S; V) + α3*E_surface_reflection, wherein α1, α2 and α3 are preset sub-weight coefficients, and the calculation result is normalized.

8. The method of claim 1, wherein the method is characterized by: The process of calculating the comprehensive aesthetic entropy value comprises: The curve entropy, the proportion entropy and the overall entropy are respectively processed by using a Min-Max normalization method, so as to map the aesthetic entropy values of each dimension to the interval [0, 1], and to obtain normalized curve entropy E'_curve, normalized proportion entropy E'_proportion and normalized overall entropy E'_total. The comprehensive aesthetic entropy value is calculated according to a formula E_comprehensive = β1*E'_curve + β2*E'_proportion + β3*E'_total, wherein β1, β2 and β3 are preset dimension weight coefficients.

9. An aesthetic entropy-based intelligent screening device for automobile styling, characterized in that, It comprises: a modeling feature extraction unit configured to acquire image data of a plurality of candidate automobile modeling schemes, and to pre-process the image data to extract modeling features of each candidate automobile modeling scheme, the modeling features comprising curve features, proportion features and color features; a multi-dimensional aesthetic entropy value calculation unit configured to calculate multi-dimensional aesthetic entropy values of each candidate automobile modeling scheme according to the modeling features, the multi-dimensional aesthetic entropy values comprising curve entropy, proportion entropy and overall entropy, wherein the curve entropy is calculated by using differential entropy to calculate the information amount of continuous curvature change, the proportion entropy is calculated by using relative entropy to calculate the deviation degree of actual proportion from ideal proportion, and the overall entropy is calculated by using joint entropy to calculate the comprehensive information amount of multiple color channels; a comprehensive aesthetic entropy value calculation unit configured to calculate a comprehensive aesthetic entropy value of each candidate automobile modeling scheme according to the multi-dimensional aesthetic entropy values; a target modeling scheme screening unit configured to screen the comprehensive aesthetic entropy values of each candidate automobile modeling scheme according to a preset screening condition, and to output a target modeling scheme meeting the condition.

10. An aesthetic entropy-based intelligent screening device for automobile styling, characterized in that, It comprises a memory and a processor, the memory stores a computer program which can be executed by the processor to implement the automobile modeling intelligent screening method based on aesthetic entropy according to any one of claims 1 to 8.

Images