A color enhancement method and system for SDR colorimeters based on HDR image priors

Abstract

This invention relates to a color enhancement method and system for an SDR colorimeter based on HDR image prior, and pertains to the field of automotive paint repair technology. The method includes: acquiring current paint repair image data in response to a paint repair signal to determine the current paint repair image data format; if the current paint repair image data format is not a standard paint repair image data format, re-determining the current paint repair image data based on format adjustment rules; determining a YUV color space image based on the current paint repair image data according to a preset transformation matrix, and extracting the luminance Y channel; performing dynamic range compression on the current paint repair image data according to the SDR standard luminance range; performing global tone consistency optimization based on the extracted paint texture and paint highlight values ​​from the current paint repair image data using a perceptual model to obtain an optimized luminance Y channel; and re-determining the current paint repair image data based on the optimized luminance Y channel. This invention improves the accuracy of paint repair color matching and enhances visual consistency with the human eye.

Description

Technical Field

[0001] This invention relates to the field of automotive paint repair technology, and in particular to a color enhancement method and system for an SDR colorimeter based on HDR image prior. Background Technology

[0002] Automotive paint touch-up and color matching are crucial aspects of automotive repair and bodywork painting, with the core being the accurate reproduction of the original color, brightness, and texture of the paint. Currently, the industry commonly uses colorimeters to capture images of the paint surface and digitally identify the paint color parameters to provide a basis for color formulation.

[0003] As imaging equipment becomes more precise, colorimeters often output high dynamic range (HDR) images to fully preserve the highlights, shadows, and texture details of vehicle paint. However, in actual display, transmission, and color correction, systems mostly use standard dynamic range (SDR) devices, thus requiring the conversion of HDR images to SDR images.

[0004] Regarding the aforementioned technologies, in the process of converting high dynamic range (HDR) to standard dynamic range (SDR) images of car paint repair based on a colorimeter, conventional dynamic range compression methods only perform simple linear or nonlinear mapping on brightness, which easily leads to the loss of key textures such as metallic particles, pearlescent textures, and orange peel textures on the car paint surface, and cannot meet the image quality requirements of car paint repair color measurement. Summary of the Invention

[0005] To address the issue of lost key textures in automotive paint and failure to meet the quality requirements of paint touch-up colorimetric images during the conversion from HDR to SDR images, this invention provides a color enhancement method and system for SDR colorimeters based on HDR image priors.

[0006] In a first aspect, the present invention provides a color enhancement method for an SDR colorimeter based on HDR image priors, employing the following technical solution:

[0007] A color enhancement method for SDR colorimeters based on HDR image priors includes:

[0008] Step 1: In response to the paint touch-up signal, acquire the current paint touch-up image data to determine the current paint touch-up image data format;

[0009] Step 2: If the current paint touch-up image data format is not the preset standard paint touch-up image data format, redetermine the current paint touch-up image data based on the preset format adjustment rules;

[0010] Step 3: Based on the current paint touch-up image data, determine the YUV color space image according to the preset transformation matrix, and extract the luminance Y channel;

[0011] Step 4: Perform dynamic range compression on the current paint touch-up image data according to the preset SDR standard brightness range;

[0012] Step 5: Based on the preset perception model, extract the paint texture and paint highlight value corresponding to the current paint touch-up image data and perform a global tone consistency optimization operation to obtain the optimized brightness Y channel;

[0013] Step 6: Redetermine the current paint touch-up image data based on the optimized brightness Y channel.

[0014] By adopting the above technical solution, when converting the paint touch-up image from HDR to SDR, the brightness channel is separated by the YUV color space and dynamic range compression is performed. Combined with the perception model, the global tone consistency of the paint texture and highlights is optimized, avoiding the problems of paint detail loss, highlight distortion and tone unevenness caused by traditional direct compression, and ensuring the restoration accuracy and visual consistency of the paint touch-up color measurement image.

[0015] Optionally, methods for performing global tone consistency optimization operations include:

[0016] Step 60: Determine the current paint type based on the current paint repair image data;

[0017] Step 61: Determine the current abnormal area of ​​the paint based on the current paint repair image data and the current paint type;

[0018] Step 62: Determine adjacent areas of the paint based on the current abnormal area;

[0019] Step 63: Determine the light interference area based on the adjacent areas of the paint and the current abnormal areas of the paint;

[0020] Step 64: When there is a light interference area, determine the attenuation coefficient of the current overexposed area and the gamma coefficient of the current underexposed area according to the preset paint type mapping table based on the current paint type;

[0021] Step 65: Perform a global tone consistency optimization operation based on the current overexposed area attenuation coefficient and the current underexposed area gamma coefficient.

[0022] By adopting the above technical solution, the light interference area can be identified according to different car paint types. The overexposure attenuation coefficient and underexposure gamma coefficient can be adaptively matched to adjust the brightness of the light-affected area and improve the naturalness and uniformity of the overall tone.

[0023] Optionally, methods for determining the current abnormal area of ​​the paint based on the current paint repair image data and the current paint type include:

[0024] Step 6250: Determine adjacent pixels of the paint and abnormal pixels of the current paint based on adjacent areas of the paint and the current abnormal areas of the paint.

[0025] Step 6251: Determine the interval distance based on the adjacent areas of the paint and the current abnormal areas of the paint;

[0026] Step 6252: Determine the single-pixel gradient based on adjacent pixels of the paint, abnormal pixels of the current paint, and the interval distance;

[0027] Step 6253: Obtain the total number of abnormal pixels corresponding to the current abnormal area of ​​the vehicle paint;

[0028] Step 6254: Calculate the brightness gradient based on the total number of abnormal pixels and the gradient of a single pixel;

[0029] Step 6255: Determine the reliable brightness gradient range based on the current paint type. When the brightness gradient does not fall within the reliable brightness gradient range, determine the interference area of ​​the abnormal object.

[0030] Step 6256: When the brightness gradient falls within the reliable brightness gradient range, determine the light interference region.

[0031] By adopting the above technical solution, the brightness gradient between the abnormal area and the adjacent area can be calculated to distinguish whether the brightness change is caused by light interference or by interference from foreign objects, thereby improving the accuracy of light interference area identification and avoiding incorrect optimization.

[0032] Optionally, methods for determining the light interference region based on adjacent areas of the vehicle paint and current abnormal areas of the vehicle paint include:

[0033] Step 620: When the paint type is not a solid color paint type, determine the texture features of the adjacent paint area and the texture features of the abnormal paint area based on the adjacent paint area and the current abnormal paint area;

[0034] Step 621: Determine texture continuity based on the texture features of adjacent areas of the current paint and the texture features of abnormal areas of the paint;

[0035] Step 622: When the texture continuity falls within the preset reliable texture continuity range, determine the light interference region.

[0036] By adopting the above technical solution, for non-solid color car paints such as metallic paint and pearlescent paint, the source of abnormal brightness is determined by the continuity of texture. Under the premise of ensuring texture consistency, only the influence of light is optimized without destroying the inherent texture of the car paint.

[0037] Optionally, it also includes a method for determining the light interference region when the texture continuity does not fall within a reliable texture continuity range or the current paint type is a solid color paint type, the method comprising:

[0038] Step 623: Determine the chromaticity of the adjacent paint area and the chromaticity of the current abnormal paint area based on the adjacent paint area and the current abnormal paint area;

[0039] Step 624: Determine the chromaticity difference by comparing the chromaticity of adjacent areas of the paint with the chromaticity of the current abnormal area of ​​the paint;

[0040] Step 625: When the chromaticity difference falls within the preset reliable chromaticity difference range, determine the light interference region;

[0041] Step 626: When the chromaticity difference does not fall within the reliable chromaticity difference range, determine the interference region of the abnormal object.

[0042] By adopting the above technical solution, for solid-color car paint without obvious texture, the light interference and foreign object interference can be distinguished by the color difference, so that the light interference area identification can be applied to all types of car paint, thus improving the versatility of the solution.

[0043] Optionally, methods for performing global tone consistency optimization to obtain an optimized luminance Y channel include:

[0044] Step 50: Construct a multi-scale difference pyramid L0, L1, ..., L based on the luminance Y channel. S S represents the number of levels in the multi-scale difference pyramid;

[0045] Step 51: Perform perceptual differential operation based on the multi-scale differential pyramid to generate the current differential pyramid;

[0046] Step 52: Perform a multi-scale fusion operation based on the current difference pyramid to obtain intermediate representation values;

[0047] Step 53: Perform a partition mapping operation based on the intermediate representation values ​​and the light interference region to obtain the partition mapping result;

[0048] Step 54: Determine the current paint touch-up image data based on the partition mapping results and intermediate representation values.

[0049] By adopting the above technical solution, the multi-scale differential pyramid is used to separate details at different levels. The micro-texture of the paint is preserved by perceptual differential and multi-scale fusion, and partitioned mapping is performed in the light interference area to achieve a balance between detail preservation and dynamic range compression.

[0050] Optional, also includes:

[0051] Step 55: Obtain the initial HDR paint touch-up image data and the current SDR paint touch-up image data;

[0052] Step 56: Calculate the perceptual difference metric between the initial HDR paint touch-up image data and the current SDR paint touch-up image data based on the preset perceptual quality metric;

[0053] Step 57: When the perceived difference measurement value exceeds the preset human eye perceived quality threshold, determine the unqualified parameter;

[0054] Step 58: Determine the corrected attention weights based on the unqualified parameters, and re-execute the multi-scale fusion operation according to the corrected attention weights to obtain intermediate representation values. Then, execute steps 53 to 57 until the perceptual difference metric does not exceed the human eye's perceptual quality threshold.

[0055] By adopting the above technical solution, image quality is verified based on the quality perceived by the human eye, and attention weights are adaptively adjusted and re-optimized according to the unqualified items, so that the final SDR image is more in line with the visual habits of the human eye and the output reliability is improved.

[0056] Optional, also includes:

[0057] Step 59: When the perceived difference metric does not exceed the human eye's perceived quality threshold, store the HDR initial paint touch-up image data and the SDR current paint touch-up image data as sample paint touch-up image import data and sample paint touch-up image result data in the preset paint touch-up history database.

[0058] Step 60: If current paint touch-up image data exists, determine the similarity based on the current paint touch-up image data and the imported data of the sample paint touch-up image;

[0059] Step 61: When the similarity falls within the preset reliable similarity range, output the current paint touch-up image data based on the sample paint touch-up image result data.

[0060] By adopting the above technical solution, qualified converted HDR and SDR images are stored in the historical database. New paint touch-up tasks can directly reuse historical results through similarity matching, reducing redundant calculations and improving color measurement and adjustment efficiency.

[0061] Optionally, an image reconstruction method may also be included, which includes:

[0062] Step 62: Real-time traversal of the paint repair history database. When there is sample paint repair image result data but no corresponding sample paint repair image import data, extract the paint type, brightness distribution, and texture features corresponding to the sample paint repair image result data as side information.

[0063] Step 63: Perform feature fusion based on the sample paint touch-up image result data and side information to obtain a fused feature map;

[0064] Step 64: Based on the fused feature map and the preset reconstruction mapping network, determine the import data of the sample paint touch-up image and update the paint touch-up history database;

[0065] Step 65: If there is imported sample paint touch-up image data but no corresponding sample paint touch-up image result data, delete the imported sample paint touch-up image data.

[0066] By adopting the above technical solution, SDR images are reconstructed into HDR images for historical samples that only have SDR image results but no HDR image data, thus completing the integrity of the database samples; at the same time, invalid and dirty data are deleted to ensure the efficient use of the database.

[0067] Secondly, the present invention provides a color enhancement system for an SDR colorimeter based on HDR image priors, employing the following technical solution:

[0068] A color enhancement system for an SDR colorimeter based on HDR image priors, comprising:

[0069] The acquisition module is used to acquire the current paint touch-up image data;

[0070] The memory is used to store the program of the SDR colorimeter color enhancement method based on HDR image prior as described above;

[0071] The processor loads and executes programs from memory.

[0072] By adopting the above technical solution, the system acquires the paint image data output by the colorimeter in real time through the acquisition module, stores all the program logic of the above color enhancement method in the memory, and the processor, as the core computing unit, sequentially executes the entire process of format verification, color space conversion, dynamic range compression, perception optimization and quality feedback, so as to realize the automation and standardization of HDR to SDR conversion, ensuring that the paint touch-up color measurement image can maintain high fidelity and consistency on various display terminals, and providing a reliable visual basis for the subsequent color formula generation.

[0073] In summary, the present invention has at least one of the following beneficial technical effects:

[0074] 1. By using YUV color space separation and global tone consistency optimization driven by the perception model, micro-details such as metal particles, pearlescent texture, and orange peel texture of car paint are effectively preserved during the HDR to SDR conversion process, solving the problems of texture loss and specular distortion caused by traditional compression methods.

[0075] 2. Based on the adaptive identification of light interference areas according to the type of car paint, and combined with multi-dimensional features such as brightness gradient, texture continuity, and color difference, the system distinguishes between light influence and foreign object interference, thereby achieving precise optimization for different car paint materials and improving the environmental adaptability and robustness of the solution.

[0076] 3. By introducing a multi-scale difference pyramid and a perceptual quality feedback mechanism, iterative optimization is performed with human visual characteristics as constraints, so that the output SDR image conforms to subjective visual habits. Through similarity matching and bidirectional reconstruction functions of historical database, a data closed loop is formed to continuously improve processing efficiency and sample completeness. Attached Figure Description

[0077] Figure 1 This is a flowchart of a color enhancement method for an SDR colorimeter based on HDR image priors, as described in an embodiment of this application. Detailed Implementation

[0078] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0079] This invention discloses a color enhancement method for SDR colorimeters based on HDR image priors. (Refer to...) Figure 1 A color enhancement method for SDR colorimeters based on HDR image priors includes:

[0080] Step 1: In response to the paint touch-up signal, acquire the current paint touch-up image data to determine the current paint touch-up image data format.

[0081] The touch-up signal is the instruction that triggers the intelligent paint color matching system to start the image acquisition and processing process, and it is triggered manually.

[0082] Current paint touch-up image data refers to the original wide dynamic range image data, or HDR image data, collected by a colorimeter or image acquisition device for the area of ​​paint to be touched up. It includes the color information, brightness distribution, metallic particle texture, and high-gloss reflection characteristics of the paint.

[0083] The current paint touch-up image data format refers to the set of technical parameters of the current paint touch-up image data, specifically including color space identifier (such as sRGB, Adobe RGB, linear RGB, XYZ, etc.), brightness dynamic range (such as HDR wide range of 0~10000nit), encoding method (such as linear encoding, Gamma encoding, logarithmic encoding) and data bit depth (such as 16-bit, 32-bit floating point).

[0084] Step 2: If the current paint touch-up image data format is not the preset standard paint touch-up image data format, redetermine the current paint touch-up image data based on the preset format adjustment rules.

[0085] The standard paint touch-up image data format refers to the input image format uniformly specified by the intelligent paint color matching system. Specifically, the color space is linear sRGB, the brightness range is linear brightness values ​​normalized to [0, 1], the encoding method is linear floating-point encoding, and the data bit depth is 32-bit floating-point number.

[0086] The format adjustment rules refer to a set of preset operations for converting non-standard paint touch-up image data into standard paint touch-up image data format. Specifically, it uses XYZ to convert to a standard sRGB matrix, which is a common technique in this field. Then, it performs Gamma correction to restore linear brightness values ​​and finally linearly scales the brightness range to the normalized range of [0, 1].

[0087] If the current paint touch-up image data format is not the standard paint touch-up image data format, it means that the color space, encoding method or brightness range of the original image does not meet the system's unified requirements. Directly using it for subsequent processing will lead to color adjustment algorithm calculation deviation, color deviation or loss of details. Therefore, it is necessary to perform format adjustment.

[0088] Step 3: Based on the current paint touch-up image data, determine the YUV color space image according to the preset transformation matrix, and extract the luminance Y channel.

[0089] The transformation matrix is ​​a matrix that converts an RGB image to the YUV color space. This matrix has fixed values, specifically... .

[0090] YUV color space images refer to images that are converted from RGB images to images with separate luminance (Y) and chrominance (U, V) channels through a transformation matrix. The Y channel contains the luminance information of the image and is decoupled from the color information; the U and V channels contain the chrominance information of the image and are decoupled from the luminance information.

[0091] The luminance Y channel refers to the single-channel data in a YUV color space image that represents the luminance distribution of the image. Its value is directly related to the brightness of each pixel in the image and is completely independent of the chromaticity information of the U and V channels.

[0092] Step 4: Perform dynamic range compression on the current paint touch-up image data according to the preset SDR standard brightness range.

[0093] The SDR standard brightness range refers to the range of brightness values ​​for a standard dynamic range image. In this embodiment, it is defined as a normalized [0,1] linear brightness interval, corresponding to the displayable brightness range of the display device.

[0094] Dynamic range compression refers to the operation of compressing the wide HDR brightness range (e.g., 0~10000 nits) to the SDR standard brightness range [0,1]. In this embodiment, only the brightness Y channel is compressed, while the U and V channels remain unchanged to ensure that the hue and saturation of the paint color do not shift.

[0095] Step 5: Based on the preset perception model, extract the paint texture and paint highlight value corresponding to the current paint touch-up image data and perform a global tone consistency optimization operation to obtain the optimized brightness Y channel.

[0096] A perception model is a mathematical model built on the human visual system (HVS) to simulate the human eye's perception characteristics of brightness, contrast, and color.

[0097] Paint texture refers to the microscopic structural features of the paint surface, including the distribution of metallic aluminum powder particles in metallic paint, the pearlescent reflective texture in pearlescent paint, and the orange peel texture of the clear coat, etc., which are the core characteristics of paint texture.

[0098] High gloss value of automotive paint refers to the numerical representation of the high-brightness area formed by the reflection of external light on the surface of automotive paint, including the local high gloss reflection value of metal particles, the specular reflection high gloss value of the clear coat layer, and the global high gloss intensity value under direct sunlight.

[0099] Global tone consistency optimization refers to the operation of adjusting the global brightness of the compressed Y channel based on a perceptual model. Specifically, it involves progressive compression of overexposed areas, local contrast enhancement of underexposed areas, and smooth tone transition in mid-brightness areas.

[0100] Optimized brightness Y channel refers to the brightness channel after extracting the paint texture and highlight value through the perception model and performing global tone consistency optimization. Its value not only preserves the details and highlight features of the paint, but also meets the naturalness requirements of human visual perception.

[0101] Step 6: Redetermine the current paint touch-up image data based on the optimized brightness Y channel.

[0102] The optimized Y channel is recombined with the initial U and V channels to form a YUV image, which is then restored to linear sRGB format SDR paint touch-up image data through the YUV matrix.

[0103] The methods for performing global tone consistency optimization operations include:

[0104] Step 60: Determine the current paint type based on the current paint repair image data.

[0105] The current paint type refers to the category of paint classified based on the current paint repair image data, specifically divided into three categories: solid color paint, metallic paint, and pearlescent paint. Solid color paint refers to a pure color paint without metallic or pearlescent particles. Its surface is smooth and the color is uniform, commonly found in basic colors such as white, black, and red. Metallic paint refers to paint with added metallic particles such as aluminum powder. Its surface has a metallic luster and a grainy feel, and it will shimmer under light. Pearlescent paint is paint with added pearlescent flakes. Its pearlescent flakes have a multi-layered structure, which can produce a soft rainbow luster through light refraction and interference. The color changes more obviously with the viewing angle.

[0106] Step 61: Determine the current abnormal area of ​​the paint based on the current paint repair image data and the current paint type.

[0107] The current abnormal paint area refers to the set of pixels extracted from the current paint repair image data using dynamic range compression whose brightness values ​​deviate from the preset normal paint brightness range. The normal paint brightness range refers to the brightness threshold range set for different paint types, obtained through statistical analysis of a large number of standard paint sample images.

[0108] Step 62: Determine the adjacent areas of the paint based on the current abnormal area of ​​the paint.

[0109] The adjacent area of ​​the paint refers to the normal paint pixel area extracted around the current abnormal paint area.

[0110] Step 63: Determine the light interference area based on the adjacent areas of the paint and the current abnormal areas of the paint.

[0111] The light interference area refers to the area of ​​abnormal brightness caused by ambient light that is selected from the current abnormal area of ​​the paint, and is used to distinguish the current abnormal area of ​​the paint caused by physical defects in the paint itself.

[0112] Step 64: When there is a light interference area, determine the attenuation coefficient of the current overexposed area and the gamma coefficient of the current underexposed area based on the current paint type according to the preset paint type mapping table.

[0113] The paint type mapping table is a reference table that is pre-constructed based on a large amount of measured data of paint images, and associates paint type with tone optimization parameters. Here, tone optimization parameters are the attenuation coefficient of the current overexposed area and the gamma coefficient of the current underexposed area.

[0114] The current overexposed area attenuation coefficient refers to the coefficient matched from the paint type mapping table based on the current paint type. It is used to perform non-linear progressive compression on overexposed pixels in the light interference area and controls the degree of brightness compression in the overexposed area.

[0115] The gamma coefficient for the current underexposed area refers to the gamma correction coefficient matched from the paint type mapping table based on the current paint type. It is used to perform local contrast enhancement on underexposed pixels in the light interference area, and its function is to improve the recognition of dark textures (such as orange peel texture) in the paint.

[0116] When there is a light interference area, it indicates that there is a brightness and darkness abnormality caused by ambient light in the current abnormal area of ​​the vehicle paint, which is a prerequisite for performing global tone consistency optimization.

[0117] Step 65: Perform a global tone consistency optimization operation based on the current overexposed area attenuation coefficient and the current underexposed area gamma coefficient.

[0118] When there is an attenuation coefficient for the current overexposed area and a gamma coefficient for the current underexposed area, it indicates that the paint type identification and exclusive optimization parameter matching have been completed, and there is a light interference area that can be optimized. At this time, differential tone optimization can be performed on the light interference area based on these two coefficients.

[0119] The methods for determining the current abnormal area of ​​the paint based on the current paint repair image data and the current paint type include:

[0120] Step 6250: Determine the adjacent pixels of the paint and the current abnormal pixels of the paint based on the adjacent areas of the paint and the current abnormal areas of the paint.

[0121] Adjacent paint pixels refer to the normal paint pixels within the adjacent area of ​​the paint that are closest to the current abnormal paint pixel, and are used to compare the brightness of the abnormal paint pixel.

[0122] The current abnormal pixel in the paint refers to a single pixel within the current abnormal area of ​​the paint.

[0123] Step 6251: Determine the interval distance based on the adjacent areas of the paint and the current abnormal areas of the paint.

[0124] The interval distance refers to the pixel-level Euclidean distance between the current abnormal pixel of the paint and the corresponding adjacent pixel of the paint. This calculation method is a common technique in this field.

[0125] Step 6252: Determine the single-pixel gradient based on adjacent pixels of the paint, abnormal pixels of the current paint, and the interval distance.

[0126] The single-pixel gradient refers to the ratio of the brightness difference between the current abnormal pixel and its adjacent pixel to the distance between them. It is calculated using the formula g. i = |Y1-Y2| / d, where g i Y1 is the brightness value corresponding to the abnormal pixel of the current car paint, Y2 is the brightness value corresponding to the adjacent pixel of the car paint, and d is the interval distance.

[0127] Step 6253: Obtain the total number of abnormal pixels corresponding to the current abnormal area of ​​the car paint.

[0128] The total number of abnormal pixels refers to the total number of valid abnormal pixels within the current abnormal area of ​​the vehicle paint.

[0129] Step 6254: Calculate the brightness gradient based on the total number of abnormal pixels and the gradient of a single pixel.

[0130] Brightness gradient refers to the arithmetic mean of the gradients of all individual pixels within the current abnormal area of ​​the vehicle paint. It is used to quantify the degree of brightness abrupt change in the entire abnormal area. The calculation formula is as follows: , where G is the brightness gradient and N is the total number of abnormal pixels.

[0131] Step 6255: Determine the reliable brightness gradient range based on the current paint type. When the brightness gradient does not fall within the reliable brightness gradient range, determine the interference area of ​​the abnormal object.

[0132] The reliable brightness gradient range refers to the brightness gradient interval corresponding to the light interference areas of different paint types, obtained based on repeated experimental measurements of paint scenarios. When the brightness gradient does not fall within the reliable brightness gradient range, it indicates that the brightness abrupt change in the current abnormal area of ​​the paint does not conform to the characteristics caused by light interference. It may be caused by abnormal objects such as foreign object obstruction or paint damage. In this case, the area needs to be marked as an abnormal object interference area to exclude it from the recognition range of the light interference area. An abnormal object interference area refers to a local brightness abnormality area caused by foreign objects (such as dust, scratches, water stains, glue stains, etc.) or external obstructions (such as leaves, fingers, etc.) on the paint surface. Its brightness characteristics are significantly different from those of the light interference area and need to be separately marked and excluded in subsequent processing to avoid interfering with the tone optimization algorithm.

[0133] Step 6256: When the brightness gradient falls within the reliable brightness gradient range, determine the light interference region.

[0134] When the brightness gradient falls within the reliable brightness gradient range, it indicates that the brightness change in the abnormal area of ​​the current paint conforms to the law of ambient light propagation and has no obvious hard boundary. Therefore, it is determined to be a light interference area and has the conditions for subsequent tone optimization.

[0135] The methods for determining the light interference region based on adjacent areas of the vehicle paint and current abnormal areas of the vehicle paint include:

[0136] Step 620: When the paint type is not a solid color paint type, determine the texture features of the adjacent paint area and the texture features of the abnormal paint area based on the adjacent paint area and the current abnormal paint area.

[0137] When the paint type is not a solid color paint type, it means that the current paint is metallic paint or pearlescent paint. Its surface has obvious micro-texture structures such as metallic particles and pearlescent flakes. The abnormal brightness can be distinguished by whether the texture features are continuous. It is caused by light interference or by physical defects such as scratches and stains.

[0138] The texture features of adjacent areas of vehicle paint refer to the feature information extracted from the normal paint surface of adjacent areas of vehicle paint to characterize the inherent microstructure of vehicle paint, including the distribution of metal particles, pearlescent reflection texture, and the position and direction of paint surface texture.

[0139] Texture features of abnormal paint areas refer to texture information extracted from the current abnormal paint area that is of the same type as the texture features of adjacent paint areas.

[0140] Step 621: Determine texture continuity based on the texture features of adjacent areas of the current paint and the texture features of abnormal areas of the paint.

[0141] Texture continuity refers to the degree of matching between the texture features of the current abnormal area of ​​the paint and the texture features of the adjacent areas of the paint. It is used to determine whether the texture of the paint surface has broken or changed abruptly.

[0142] Step 622: When the texture continuity falls within the preset reliable texture continuity range, determine the light interference region.

[0143] When the texture continuity falls within the reliable texture continuity range, it means that the texture inside the abnormal area of ​​the current paint is consistent with the normal paint texture. The paint itself has no physical damage or foreign matter covering it, and its brightness change is caused by ambient light. Therefore, it is determined to be a light interference area.

[0144] This includes a method for determining the light interference region when the texture continuity does not fall within a reliable texture continuity range or the current paint type is a solid color paint type. This method includes:

[0145] Step 623: Determine the chromaticity of the adjacent paint area and the chromaticity of the current abnormal paint area based on the adjacent paint area and the current abnormal paint area.

[0146] The chromaticity of adjacent areas of vehicle paint refers to the chromaticity information extracted from the normal paint surface of adjacent areas of the vehicle paint.

[0147] The current abnormal paint area chromaticity refers to the color information extracted from the current abnormal paint area that is of the same dimension as the chromaticity of the adjacent paint area, and is used for comparison with the normal paint chromaticity.

[0148] Step 624: Determine the chromaticity difference by comparing the chromaticity of adjacent areas of the paint with the chromaticity of the current abnormal area of ​​the paint.

[0149] The chromaticity difference value refers to the difference between the chromaticity of adjacent areas of the paint and the chromaticity of the current abnormal area of ​​the paint, and is used to determine whether a color change has occurred in the abnormal area.

[0150] Step 625: When the chromaticity difference falls within the preset reliable chromaticity difference range, determine the light interference region.

[0151] The reliable chromaticity difference range refers to the range of minute chromaticity differences allowed by light interference, determined based on automotive paint industry standards and a large amount of measured data.

[0152] When the chromaticity difference value falls within the reliable chromaticity difference value range, it indicates that the abnormal area of ​​the current car paint only has a change in brightness and no obvious color change, and its brightness abnormality is caused by ambient light.

[0153] Step 626: When the chromaticity difference does not fall within the reliable chromaticity difference range, determine the interference region of the abnormal object.

[0154] When the chromaticity difference does not fall within the reliable chromaticity difference range, it indicates that there is a significant color change in the abnormal area of ​​the current car paint. This is not simply due to the influence of light, but is caused by abnormal objects such as scratches, stains, and foreign objects.

[0155] The methods for performing global tone consistency optimization to obtain optimized luminance Y channel include:

[0156] Step 50: Construct a multi-scale difference pyramid L0, L1, ..., L based on the luminance Y channel. S S represents the number of levels in the multi-scale difference pyramid.

[0157] Multiscale difference pyramids refer to multi-layer image structures formed by downsampling the dynamic range-compressed luminance Y channel layer by layer according to different sizes, which are used to extract texture, highlight and brightness distribution information of different sizes in car paint images.

[0158] The number of layers in a multi-scale difference pyramid refers to the total number of image layers contained in the pyramid. It controls the fineness of texture extraction. For example, the bottom layer L0 corresponds to a 1000×1000 pixel metallic paint image. This layer clearly shows each metal particle, exhibiting rich detail. In this example, where S is 3, L1 performs 2×2 average pooling downsampling on L0, merging four pixels into one, resulting in a 500×500 pixel image. This layer has fewer metal particle details and is smoother overall. Finally, L2 performs 2×2 average pooling downsampling to obtain a 250×250 pixel image. This layer shows almost no metal particles, preserving only the overall brightness and shadows of the image.

[0159] Step 51: Perform perceptual differential operation based on the multi-scale differential pyramid to generate the current differential pyramid.

[0160] Perceptual differential operation refers to the operation of calculating the brightness difference between adjacent layers of a multi-scale differential pyramid, and separating high-frequency details such as paint texture, highlights, and orange peel texture at different scales from the overall brightness.

[0161] The current differential pyramid refers to a multi-layer differential data set obtained through perceptual differential operations, which retains only detailed information about the paint at each scale.

[0162] Step 52: Perform a multi-scale fusion operation based on the current difference pyramid to obtain intermediate representation values.

[0163] Multi-scale fusion refers to the process of weighting and merging detailed information at different scales in the current difference pyramid according to preset attention weights, thereby unifying fine-grained highlights, medium textures, and coarse-grained shadows. Attention weights are detail weight allocation coefficients obtained through multiple experiments and adapted to different paint types. For metallic paint, fine-grained highlights (bottom layer of the difference pyramid) are assigned an attention weight of 0.7, medium textures (middle layer of the difference pyramid) are assigned an attention weight of 0.2, and coarse-grained shadows (top layer of the difference pyramid) are assigned an attention weight of 0.1.

[0164] Intermediate representation values ​​refer to intermediate image data that contains complete details of the vehicle paint, obtained after multi-scale fusion.

[0165] Step 53: Perform a partition mapping operation based on the intermediate representation value and the light interference region to obtain the partition mapping result.

[0166] Partition mapping refers to the process of dividing an image into overexposed, underexposed, and normal regions based on the identified light interference areas, and then using corresponding attenuation and gamma coefficients to perform differentiated brightness adjustments for each region.

[0167] The partition mapping result refers to the brightness channel data that has consistent hue, no excessive reflection, and no loss of detail in dark areas after partition-differentiated brightness adjustment.

[0168] Step 54: Determine the current paint touch-up image data based on the partition mapping results and intermediate representation values.

[0169] When there are partition mapping results and intermediate representation values, it means that the paint detail preservation and global tone consistency adjustment have been completed. The final optimized brightness Y channel can be output, and the current paint repair image data can be determined.

[0170] This also includes:

[0171] Step 55: Obtain the initial HDR paint touch-up image data and the current SDR paint touch-up image data.

[0172] HDR initial touch-up image data refers to high dynamic range touch-up image data directly acquired by the colorimeter after responding to the touch-up signal, without format adjustment or dynamic range compression. SDR current touch-up image data refers to touch-up image data initially determined through partition mapping results and intermediate representation values.

[0173] Step 56: Calculate the perceptual difference metric between the initial HDR paint touch-up image data and the current SDR paint touch-up image data based on the preset perceptual quality metric.

[0174] Perceived quality metric refers to a quality evaluation index that is customized for automotive paint repair scenarios and simulates the human visual system (HVS). Specifically, it adopts the normalized Laplacian pyramid distance and can be decomposed into a weighted calculation rule for multiple sub-items such as contrast retention, detail retention, color fidelity, and perceptual consistency.

[0175] Contrast retention refers to the degree to which the current SDR paint touch-up image data and the initial HDR paint touch-up image data maintain the local brightness contrast relationship. Detail retention refers to the degree to which the current SDR paint touch-up image data retains the inherent micro-details of the paint, such as metallic highlights, pearlescent reflections, and orange peel texture. Color fidelity refers to the consistency of color attributes between the current SDR paint touch-up image data and the initial HDR paint touch-up image data, calculated using the automotive paint industry standard ΔEab color difference formula. Perceptual consistency refers to whether the brightness and hue distribution of the current SDR paint touch-up image data is natural across the entire image.

[0176] The perceptual difference metric is a quantitative value calculated by perceptual quality measurement, representing the degree of difference between the initial HDR paint touch-up image data and the current SDR paint touch-up image data at the human eye level. The value ranges from [0,1]. The larger the value, the more obvious the perceptual difference and the less the image quality meets the human eye's evaluation standards.

[0177] Step 57: When the perceived difference measurement value exceeds the preset human eye perceived quality threshold, determine the unqualified parameter.

[0178] The human eye perceived quality threshold refers to the critical value of the perceived difference measure determined through a large number of subjective evaluation experiments using human eyes on car paint touch-up images.

[0179] Unqualified parameters refer to parameters that cause the perceptual difference metric to exceed the threshold. Specifically, these are the attention weights initially set in the multi-scale fusion operation, which correspond to the weight allocation coefficients for fine-grained highlights, medium textures, and coarse-grained brightness and darkness.

[0180] When the perceived difference metric exceeds the human eye's perceived quality threshold, it indicates that the current SDR paint touch-up image data does not meet the human eye's subjective evaluation standards in terms of contrast preservation, paint detail retention, or overall tone consistency. This is because the attention weight allocation of multi-scale fusion is unreasonable, resulting in the failure to effectively preserve key paint details or an overall imbalance between light and dark.

[0181] Step 58: Determine the corrected attention weights based on the unqualified parameters, and re-execute the multi-scale fusion operation according to the corrected attention weights to obtain intermediate representation values. Then, execute steps 53 to 57 until the perceptual difference metric does not exceed the human eye's perceptual quality threshold.

[0182] Corrected attention weights refer to the new weight allocation coefficients obtained by adaptively adjusting the initial attention weights based on the multi-dimensional sub-item results of the perceptual difference metric. For example, when the detail retention sub-item is low, the attention weight of fine-grained highlights is increased; when the perceptual consistency sub-item is low, the attention weight of coarse-grained brightness and darkness is increased.

[0183] This also includes an image reconstruction method, which includes:

[0184] Step 62: Real-time traversal of the paint repair history database. When there is sample paint repair image result data but no corresponding sample paint repair image import data, extract the paint type, brightness distribution, and texture features corresponding to the sample paint repair image result data as side information.

[0185] Paint type refers to the specific type of paint corresponding to the sample paint repair image result data. Brightness distribution refers to the distribution of brightness values ​​in the entire image range of the sample paint repair image result data, including the brightness values ​​and distribution ratios of highlights, shadows, and midtones. Texture features refer to the microscopic texture information of the paint extracted from the sample paint repair image result data. Side information refers to the supplementary information set assisting SDR to HDR reconstruction. This side information also includes historical optimization parameters; specifically, it needs to be verified whether the paint repair history database stores historical optimization parameters. If historical optimization parameters exist, they are added to the side information.

[0186] When sample paint touch-up image result data exists but its corresponding sample paint touch-up image import data does not exist, it means that the paint touch-up history database only stores the SDR format result image of the sample and does not store its corresponding original HDR import image. This may be due to data overwriting and loss. The HDR image needs to be restored through reconstruction to meet the needs of subsequent similarity matching.

[0187] Step 63: Perform feature fusion based on the sample paint touch-up image result data and side information to obtain a fused feature map.

[0188] Fusion feature maps refer to multi-dimensional image features extracted by convolution of sample paint touch-up image data, which are then combined with auxiliary features obtained by feature encoding of lateral information in the channel dimension to form a multi-channel feature map. This is a common technique in the field of image reconstruction.

[0189] Step 64: Based on the fused feature map and the preset reconstruction mapping network, determine the import data of the sample paint touch-up image and update the paint touch-up history database.

[0190] The reconstruction mapping network refers to the use of mature encoder-decoder convolutional neural networks (Encoder-Decoder CNN) in existing technologies. Specifically, it can use publicly available general image reconstruction network structures such as U-Net and ResNet to restore the fused feature map corresponding to the SDR image into a high dynamic range image.

[0191] Step 65: If there is imported sample paint touch-up image data but no corresponding sample paint touch-up image result data, delete the imported sample paint touch-up image data.

[0192] When there is imported sample paint touch-up image data but no corresponding sample paint touch-up image result data, it means that the paint touch-up history database only saves the original HDR input image, and the SDR output image may have been lost. Even if a new paint touch-up image is successfully matched with the sample, the effective SDR image result cannot be reused, and the optimization process still needs to be executed from scratch. Therefore, this data has no historical reuse value. Thus, this data has no historical matching and reuse value, so a cleanup and deletion operation is performed to improve storage space utilization.

[0193] Based on the same inventive concept, embodiments of the present invention provide a color enhancement system for an SDR colorimeter based on HDR image priors.

[0194] A color enhancement system for an SDR colorimeter based on HDR image priors, comprising:

[0195] The acquisition module is used to acquire the current paint touch-up image data;

[0196] The memory is used to store a program for a color enhancement method for an SDR colorimeter based on HDR image priors;

[0197] The processor loads and executes programs from memory.

[0198] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for SDR colorimeter color enhancement based on HDR image priors, characterized in that, include: Step 1: In response to the paint touch-up signal, acquire the current paint touch-up image data to determine the current paint touch-up image data format; Step 2: If the current paint touch-up image data format is not the preset standard paint touch-up image data format, redetermine the current paint touch-up image data based on the preset format adjustment rules; Step 3: Based on the current paint touch-up image data, determine the YUV color space image according to the preset transformation matrix, and extract the luminance Y channel; Step 4: Perform dynamic range compression on the current paint touch-up image data according to the preset SDR standard brightness range; Step 5: Based on the preset perception model, extract the paint texture and paint highlight value corresponding to the current paint touch-up image data and perform a global tone consistency optimization operation to obtain the optimized brightness Y channel; Step 6: Redetermine the current paint touch-up image data based on the optimized brightness Y channel; The methods for performing global tone consistency optimization operations include: Step 60: Determine the current paint type based on the current paint repair image data; Step 61: Determine the current abnormal area of ​​the paint based on the current paint repair image data and the current paint type; Step 62: Determine adjacent areas of the paint based on the current abnormal area; Step 63: Determine the light interference area based on the adjacent areas of the paint and the current abnormal areas of the paint; Among them, the light interference area refers to the area of ​​abnormal brightness caused by ambient light that is selected from the current abnormal area of ​​the paint, and is used to distinguish the current abnormal area of ​​the paint caused by physical defects in the paint itself. Step 64: When there is a light interference area, determine the attenuation coefficient of the current overexposed area and the gamma coefficient of the current underexposed area according to the preset paint type mapping table based on the current paint type; Step 65: Perform a global tone consistency optimization operation based on the current overexposed area attenuation coefficient and the current underexposed area gamma coefficient.

2. The SDR colorimeter color enhancement method based on HDR image priors of claim 1, wherein, Methods for determining the current abnormal area of ​​the paint based on the current paint repair image data and the current paint type include: Step 6250: Determine adjacent pixels of the paint and abnormal pixels of the current paint based on adjacent areas of the paint and the current abnormal areas of the paint. Step 6251: Determine the interval distance based on the adjacent areas of the paint and the current abnormal areas of the paint; Step 6252: Determine the single-pixel gradient based on adjacent pixels of the paint, abnormal pixels of the current paint, and the interval distance; Among them, the single pixel gradient refers to the ratio of the brightness difference between the current abnormal pixel of the paint and the adjacent pixel of the paint to the interval distance. Step 6253: Obtain the total number of abnormal pixels corresponding to the current abnormal area of ​​the vehicle paint; Step 6254: Calculate the brightness gradient based on the total number of abnormal pixels and the gradient of a single pixel; Among them, the brightness gradient refers to the arithmetic mean of the gradients of all single pixels in the current abnormal area of ​​the paint, which is used to quantify the degree of brightness change in the entire abnormal area. Step 6255: Determine the reliable brightness gradient range based on the current paint type. When the brightness gradient does not fall within the reliable brightness gradient range, determine the interference area of ​​the abnormal object. Step 6256: When the brightness gradient falls within the reliable brightness gradient range, determine the light interference region.

3. The SDR colorimeter color enhancement method based on HDR image priors of claim 1, wherein, Methods for determining the light interference region based on adjacent areas of the vehicle paint and current abnormal areas of the vehicle paint include: Step 620: When the paint type is not a solid color paint type, determine the texture features of the adjacent paint area and the texture features of the abnormal paint area based on the adjacent paint area and the current abnormal paint area; Step 621: Determine texture continuity based on the texture features of adjacent areas of the current paint and the texture features of abnormal areas of the paint; Step 622: When the texture continuity falls within the preset reliable texture continuity range, determine the light interference region.

4. The SDR colorimeter color enhancement method based on HDR image priors of claim 3, wherein, It also includes a method for determining the light interference region when the texture continuity does not fall within a reliable texture continuity range or the current paint type is a solid color paint type, the method including: Step 623: Determine the chromaticity of the adjacent paint area and the chromaticity of the current abnormal paint area based on the adjacent paint area and the current abnormal paint area; Step 624: Determine the chromaticity difference by comparing the chromaticity of adjacent areas of the paint with the chromaticity of the current abnormal area of ​​the paint; Step 625: When the chromaticity difference falls within the preset reliable chromaticity difference range, determine the light interference region; Step 626: When the chromaticity difference does not fall within the reliable chromaticity difference range, determine the interference region of the abnormal object.

5. The SDR colorimeter color enhancement method based on HDR image priors of claim 1, wherein, Methods for performing global tone consistency optimization to obtain an optimized luminance Y channel include: Step 50: Constructing multi-scale difference pyramid L0, L1, …, Ls based on the luminance Y channel S , S represents the number of multi-scale difference pyramid layers; Step 51: Perform perceptual differential operation based on the multi-scale differential pyramid to generate the current differential pyramid; Step 52: Perform a multi-scale fusion operation based on the current difference pyramid to obtain intermediate representation values; Step 53: Perform a partition mapping operation based on the intermediate representation values ​​and the light interference region to obtain the partition mapping result; Among them, the partition mapping operation refers to dividing the image into overexposed areas, underexposed areas and normal areas based on the identified light interference areas, and then using the corresponding attenuation coefficient and gamma coefficient to perform differentiated brightness adjustment. The partition mapping result refers to the brightness channel data that, after partition-differentiated brightness adjustment, has consistent hue, no excessive reflection, and no loss of detail in dark areas; Step 54: Determine the current paint touch-up image data based on the partition mapping results and intermediate representation values.

6. The SDR colorimeter color enhancement method based on HDR image priors of claim 5, wherein, Also includes: Step 55: Obtain the initial HDR paint touch-up image data and the current SDR paint touch-up image data; Step 56: Calculate the perceptual difference metric between the initial HDR paint touch-up image data and the current SDR paint touch-up image data based on the preset perceptual quality metric; Step 57: When the perceived difference measurement value exceeds the preset human eye perceived quality threshold, determine the unqualified parameter; Step 58: Determine the corrected attention weights based on the unqualified parameters, and re-execute the multi-scale fusion operation according to the corrected attention weights to obtain intermediate representation values. Then, execute steps 53 to 57 until the perceptual difference metric does not exceed the human eye's perceptual quality threshold.

7. The SDR colorimeter color enhancement method based on HDR image priors of claim 6, wherein, Also includes: Step 59: When the perceived difference metric does not exceed the human eye's perceived quality threshold, store the HDR initial paint touch-up image data and the SDR current paint touch-up image data as sample paint touch-up image import data and sample paint touch-up image result data in the preset paint touch-up history database. Step 60: If current paint touch-up image data exists, determine the similarity based on the current paint touch-up image data and the imported data of the sample paint touch-up image; Step 61: When the similarity falls within the preset reliable similarity range, output the current paint touch-up image data based on the sample paint touch-up image result data.

8. The SDR colorimeter color enhancement method based on HDR image priors of claim 7, wherein, It also includes image reconstruction methods, which include: Step 62: Real-time traversal of the paint repair history database. When there is sample paint repair image result data but no corresponding sample paint repair image import data, extract the paint type, brightness distribution, and texture features corresponding to the sample paint repair image result data as side information. Step 63: Perform feature fusion based on the sample paint touch-up image result data and side information to obtain a fused feature map; Step 64: Based on the fused feature map and the preset reconstruction mapping network, determine the import data of the sample paint touch-up image and update the paint touch-up history database; Step 65: If there is imported sample paint touch-up image data but no corresponding sample paint touch-up image result data, delete the imported sample paint touch-up image data.

9. An SDR colorimeter color enhancement system based on HDR image priors, characterized in that, include: The acquisition module is used to acquire the current paint touch-up image data; A memory for storing a program for a color enhancement method for an SDR colorimeter based on HDR image priors as described in any one of claims 1 to 8; The processor loads and executes programs from memory.

Images