Digital museum collection visual enhancement method based on feature fusion

Abstract

The present application relates to the field of image processing, especially to a digital museum collection visual enhancement method based on feature fusion. The subject-cut collection image is acquired and converted to a color space, and the highlight candidate region and its peripheral local reference ring are determined according to the luminance component. The luminance layering is performed on the highlight candidate region, the luminance increase degree and the saturation decrease degree are determined according to the luminance statistics and the saturation statistics of two groups of pixels, and the judgment is combined with the saturation fluctuation degree of the local reference ring. The specular highlight region is determined according to the gradient reduction degree of the same luminance layering and the gradient fluctuation degree of the local reference ring. The local correction is performed according to the luminance increase degree and the saturation decrease degree, the corrected data is converted to data, and the original data is reserved for the pixels without local correction, and the visual enhancement image is obtained. The present application can reduce the possibility of real light color region being mistakenly corrected as specular highlight.

Description

Technical Field

[0001] This invention relates to the field of image processing, and more particularly to a method for visual enhancement of digital museum collections based on feature fusion. Background Technology

[0002] Digital museums typically acquire two-dimensional color images of their collections, such as ceramics, jade, lacquerware, gold and silver artifacts, and bronzes, using digital cameras, professional photographic equipment, or scanning imaging devices. These images are then used for digital archives, web displays, and magnified viewing of specific areas. For artifacts with glazes, polished layers, or metallic reflective surfaces, lighting can easily create localized specular highlights in the acquired images, increasing the brightness of the corresponding areas and weakening original color differences and texture grayscale variations. Current image processing methods usually begin by filtering highlight areas based on their brightness, low saturation, or overexposure, followed by brightness suppression, color adjustment, or localized restoration.

[0003] For example, Chinese invention patent CN116883277B discloses a method for removing highlights from ceramic artifact images based on a dual attention mechanism and detail restoration. It reads the highlight image of the ceramic artifact, crops it, extracts features using a pre-trained VGG-19 network, combines the extracted features with the original highlight image, and then processes it through an attention module, residual structure, and detail restoration module to output a highlight-free image. The patent also points out that pixels with low saturation may be misclassified as highlight pixels, and traditional methods struggle to distinguish between pure white or near-white areas and highlight areas.

[0004] For images of digital museum collections, areas such as light-colored jade, light-colored glazes, and light-colored patterns can inherently possess high brightness, and some light-colored areas may also have low saturation. Therefore, the cause of bright areas cannot be determined solely based on a single brightness or absolute saturation. Directly including such areas in the highlight processing may alter the original brightness, color, or texture boundaries of the actual light-colored areas of the collection. Thus, a visual enhancement method is needed that can utilize the color variation relationships between different brightness levels within the candidate highlight areas and combine this with texture variations for verification. Summary of the Invention

[0005] To address the issue that both specular highlights and the light-colored materials of the artifacts themselves may appear as localized highlights in digital museum images, and that existing methods rely solely on absolute brightness or absolute saturation to determine the source of these highlights, potentially misinterpreting actual light-colored patterns or materials as specular highlights, this invention provides the following technical solution.

[0006] Visual enhancement methods for digital museum collections based on feature fusion include: Obtain the subject-cropped Collection images and conversion In a color space, candidate bright areas are determined based on luminance components, and a local reference ring around the candidate bright areas is obtained. For the bright candidate region with local reference ring, divide it into low brightness pixel group and high brightness pixel group according to brightness. Based on the brightness statistics and saturation statistics of the two groups of pixels, determine the degree of brightness increase and saturation decrease of the high brightness pixel group relative to the low brightness pixel group. Combine the degree of saturation fluctuation of local reference ring to screen the bright candidate region that meets the condition of brightness increase accompanied by saturation decrease. For a bright candidate region that meets the condition of increased brightness accompanied by decreased saturation, the pixel gradient magnitude is calculated based on the brightness component, and the gradient reduction degree of the high brightness pixel group relative to the low brightness pixel group is determined based on the gradient magnitude statistics of the two groups of pixels. The specular highlight region is determined by combining the gradient fluctuation degree of the corresponding local reference ring. The degree of local correction is determined based on the increase in brightness and decrease in saturation corresponding to the specular highlight area. The brightness and saturation components of the specular highlight area are then corrected according to this degree of local correction. Data conversion Data, and retain the original pixels without local correction. The data is used to obtain visually enhanced images.

[0007] Preferably, when the increase in brightness is positive and the decrease in saturation exceeds the saturation fluctuation of the corresponding local reference ring, the bright candidate region is determined to meet the condition of increased brightness accompanied by decreased saturation. When the gradient reduction is positive and exceeds the gradient fluctuation of the corresponding local reference ring, the bright candidate region that satisfies the condition of increased brightness accompanied by decreased saturation is determined as the specular highlight region.

[0008] Preferably, the artifact image is an eight-bit, three-channel artifact image acquired under white light illumination and with the main subject cropped; the artifact image is then converted to... After color space is applied, the luminance and saturation components are normalized to 0 to 1; The process of determining the candidate region for brightness includes: sorting the brightness components, taking the upper quartile of the brightness components as the candidate region for brightness, dividing adjacent pixels with brightness not lower than the candidate region for brightness into connected regions according to the eight-neighbor connectivity relationship, and determining connected regions with no less than four pixels as candidate regions for brightness.

[0009] Preferably, morphological dilation is performed on each bright candidate region to extend the boundary of the bright candidate region outward by 5 pixels, and the newly added adjacent pixels that do not belong to other bright candidate regions are used as the local reference ring of the bright candidate region. For candidate bright areas where local reference ring pixels cannot be obtained, retain the corresponding original image data and do not proceed to subsequent specular highlight confirmation.

[0010] Preferably, the pixels in each bright candidate region are sorted from smallest to largest according to their normalized brightness values. The pixels in the first half of the sorted region are determined as the low brightness pixel group, and the pixels in the second half of the sorted region are determined as the high brightness pixel group. When the bright candidate region contains an odd number of pixels, the pixels in the middle sorted position are assigned to the low brightness pixel group. The brightness level difference is obtained by subtracting the brightness median of the low brightness pixel group from the brightness median of the high brightness pixel group, and the brightness level difference is used as the degree of brightness increase; the saturation decrease is obtained by subtracting the saturation median of the high brightness pixel group from the saturation median of the low brightness pixel group, and the saturation decrease is used as the degree of saturation reduction. The saturation interquartile range is calculated based on the saturation data in the local reference ring corresponding to the bright candidate region, and the saturation interquartile range is used as the degree of saturation fluctuation.

[0011] Preferably, the Sobel operator is used to calculate the gradient magnitude of each pixel in the brightness component of the collection image; For a bright candidate region that satisfies the condition of increased brightness accompanied by decreased saturation, the median gradient magnitude of the corresponding pixels in the low-brightness pixel group and the high-brightness pixel group are calculated respectively. The gradient decrease amount is obtained by subtracting the median gradient magnitude of the high-brightness pixel group from the median gradient magnitude of the low-brightness pixel group, and the gradient decrease amount is used as the degree of gradient reduction. The gradient interquartile range is calculated based on the gradient magnitude of each pixel in the local reference ring corresponding to the bright candidate region, and the gradient interquartile range is used as the gradient fluctuation degree.

[0012] Preferably, the degree of local correction is determined by the brightness reduction reference amount, the saturation compensation reference amount, and the correction ratio of the pixel to be corrected; for each specular highlight area, its brightness level difference is used as the brightness reduction reference amount, its saturation decrease is used as the saturation compensation reference amount, and pixels whose normalized brightness is higher than the median brightness of the corresponding low brightness pixel group are determined as pixels to be corrected. For the pixel to be corrected, the brightness difference is obtained by subtracting the median brightness of the corresponding low-brightness pixel group from the normalized brightness of the pixel to be corrected. The correction denominator is obtained by subtracting the median brightness of the low-brightness pixel group from 1 and adding a preset positive number to make the correction denominator greater than 0. The brightness difference is divided by the correction denominator, and the resulting ratio is limited to the range of 0 to 1 to obtain the correction ratio.

[0013] Preferably, the product of the correction ratio and the brightness reduction reference amount is used as the brightness reduction amount. The brightness reduction amount is subtracted from the normalized brightness of the pixel to be corrected before correction, and the resulting value is limited to the range of 0 to 1 to obtain the normalized brightness of the pixel to be corrected after correction.

[0014] Preferably, when the normalized saturation of the pixel to be corrected is greater than 0, the product of the correction ratio and the saturation compensation reference amount is added to the normalized saturation before correction to obtain the initial saturation correction value; the larger value between the normalized saturation before correction and the median saturation of the corresponding low-brightness pixel group is taken as the upper limit of saturation, and the smaller value between the initial saturation correction value and the upper limit of saturation is taken as the normalized saturation after correction, while keeping the hue component unchanged; When the normalized saturation of the pixel to be corrected is equal to 0 before correction, the normalized saturation after correction is set to 0, and the hue component remains unchanged, only the brightness is corrected.

[0015] The present invention has the following effects: 1. Compared with the processing method of determining the specular highlight area based solely on absolute brightness or absolute low saturation, the present invention first divides the pixels into layers according to brightness within the same candidate bright area, and then compares the saturation changes between the high brightness pixel group and the low brightness pixel group, so that the saturation judgment is directly attached to the brightness change process, thereby reducing the possibility of the collection area being identified as a specular highlight area simply because the color of the collection area itself is light. 2. This invention only performs gradient decay verification on high-brightness candidate regions that meet the condition of increasing brightness accompanied by decreasing saturation, and the gradient verification still uses the same brightness layer, so that the gradient data is used to verify the aforementioned color change relationship, rather than being superimposed with brightness and saturation with fixed weights. The saturation fluctuation degree and gradient fluctuation degree of the local reference ring are used to reflect the actual color fluctuation and texture fluctuation of the adjacent artifact surface, so that the judgment benchmark changes with the local area of ​​the current artifact; wherein, the first embodiment uses the interquartile range to characterize the fluctuation degree, and the second embodiment uses the standard deviation to characterize the fluctuation degree; 3. After determining the specular highlight area, the present invention determines the pixel to be corrected and sets the correction ratio based on the brightness position of the pixel in the specular highlight area relative to the brightness statistics of the corresponding low brightness pixel group. This allows the higher brightness position to obtain a larger correction range, while retaining the original color data of the non-spectral highlight area and the pixels in the specular highlight area that have not met the correction conditions, thereby reducing the modification of the original color data of the lower brightness position in the specular highlight area. Attached Figure Description

[0016] Figure 1 This is a flowchart of steps S1 to S4 in the digital museum collection visual enhancement method based on feature fusion according to an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0018] The feature fusion described in this embodiment refers to using the brightness layers of the same highlight candidate region as a common data basis. First, a judgment is made based on the saturation changes between different brightness levels. Then, the regions that pass the aforementioned judgment are verified based on the gradient changes between the same brightness layers. After determining the specular highlight region, further local corrections are made using the brightness increase and saturation decrease obtained from the same brightness layers. This processing relationship differs from directly adding multiple independent features with fixed weights.

[0019] In this invention, luminance statistics, saturation statistics, and gradient amplitude statistics are used to characterize the overall level of luminance components, saturation components, and gradient amplitude in corresponding pixel groups; saturation fluctuation and gradient fluctuation are used to characterize the local dispersion of saturation data and gradient amplitude data in the local reference ring. For the same high-brightness candidate region, the low-brightness pixel group and the high-brightness pixel group obtain corresponding statistics using the same type of statistical rules, and the degree of brightness increase, saturation decrease, and gradient decrease when changing from a low-brightness level to a high-brightness level is determined based on the difference between the two sets of statistics. The fluctuation degree of the local reference ring serves as a local reference benchmark for the normal color or texture fluctuation of the surface of the artifacts spatially adjacent to the candidate region, in order to determine whether the color and texture changes generated by the change of brightness level within the candidate region exceed the local fluctuation range of the adjacent artifact surfaces.

[0020] Reference Figure 1 The method for visually enhancing digital museum collections based on feature fusion includes steps S1 to S4, as detailed below: S1: Acquire and preprocess the collection image to obtain the highlighted candidate region and local reference ring;

[0021] It should be noted that the input image of the artifact in this embodiment is an image of the artifact that has already been cropped, so that large areas of white background, display background, or shooting background are not included in the highlight candidate benchmark statistics. If the background area is included in the brightness ranking of the whole image, when the background brightness is higher than the specular highlight brightness of the artifact surface, it may raise the highlight candidate benchmark corresponding to the upper quartile, so that some actual specular highlights cannot enter the subsequent candidate area processing.

[0022] This embodiment uses white light illumination to ensure that when the specular reflection on the surface of the artifact is enhanced, the three color channels are mainly improved synchronously. This makes it easier to determine whether the candidate area has a color change relationship corresponding to specular highlights based on the saturation changes between different brightness levels.

[0023] Specifically, an eight-bit, three-channel image of the artifact, acquired under white light illumination and with the main body of the artifact already cropped, is read. The image is then converted to a known color space using a publicly known color space conversion method. The color space is used, and the luminance and saturation components are normalized to a range of 0 to 1. The luminance components in the main image of the artifact are sorted, and the upper quartile of the luminance component is taken as the candidate highlight criterion for the current artifact image. Adjacent pixels with luminance not lower than the upper quartile are divided into connected regions according to the eight-neighbor connectivity relationship, and connected regions with no less than four pixels are determined as candidate highlight regions.

[0024] Using no fewer than four pixels as the minimum number of pixels for the highlight candidate area ensures that both the subsequent low-brightness pixel group and the high-brightness pixel group can obtain actual pixel data, while reducing the number of isolated highlight pixels entering subsequent processing.

[0025] For each highlight candidate region, perform morphological dilation to extend the boundary of the highlight candidate region outward by 5 pixels. The newly added adjacent pixels that do not belong to other highlight candidate regions are used as the local reference ring of the highlight candidate region.

[0026] The local reference ring is located outside the highlight candidate region and is used to obtain saturation and gradient reference data that are spatially adjacent to the candidate region. For highlight candidate regions where the local reference ring pixels cannot be obtained, since reference data for local saturation and gradient fluctuations cannot be obtained, the corresponding original image data is retained and not proceeded to subsequent specular highlight confirmation.

[0027] At this point, step S1 is complete, and the data is retained. The hue component in the color space is used for subsequent local correction and color space inverse conversion. At the same time, it outputs the normalized luminance component, the normalized saturation component, the highlight candidate region for subsequent processing, and the local reference ring corresponding to the highlight candidate region.

[0028] S2: Determine whether the candidate bright areas exhibit a relationship of increased brightness accompanied by decreased saturation based on brightness layering; It should be noted that while light-colored patterns, glazes, or materials on a collectible may exhibit high brightness, this brightness is primarily determined by the collectible's inherent color properties. Therefore, pixels of varying brightness within the collectible may not simultaneously show a significant decrease in saturation. In contrast, when the surface of a collectible is subject to strong specular reflection, while the brightness of local pixels increases, the original color differences may be weakened, resulting in a more pronounced decrease in saturation for higher-brightness pixels compared to lower-brightness pixels.

[0029] Specifically, the first Pixels within a candidate region of brightness are sorted in ascending order of their normalized brightness values ​​and then divided into low-brightness and high-brightness pixel groups based on their sorting position. Pixels in the first half of the sorted region are designated as the low-brightness pixel group, and those in the second half as the high-brightness pixel group. When a candidate region contains an odd number of pixels, pixels in the middle of the sorted region are assigned to the low-brightness pixel group. Dividing the region by sorting position avoids situations where multiple pixels with the same brightness value exist within the candidate region, preventing one pixel group from lacking actual pixel data due to using the median brightness value as the dividing line.

[0030] Brightening an area does not necessarily mean a decrease in saturation. For a normalized pixel, let the maximum value of the three color channels be... The minimum value is ,exist When it is not 0, Saturation in color space and brightness satisfy: ; in, Indicates the pixel in Saturation in a color space; Indicates the pixel in Brightness in a color space; Indicates normalization The maximum value among the three channels; Indicates normalization The minimum value among the three channels.

[0031] If the three color channels are increased by the same proportion without truncation, the maximum and minimum values ​​change in the same proportion, and the aforementioned proportional relationship remains unchanged. Therefore, the saturation can remain unchanged. Thus, this embodiment does not assume that an increase in brightness necessarily leads to a decrease in saturation.

[0032] When the surface of the artifact is illuminated by near-white light and produces a strong specular reflection, from In terms of pixel value changes, the pixel values ​​of the red, green, and blue channels usually increase simultaneously, relatively reducing the original numerical differences between the three color channels. Without considering channel truncation, let's assume all three channels increase by the same positive value. Then the changed saturation and brightness satisfy: ; in, This represents the saturation obtained by simultaneously increasing the brightness of all three color channels. This represents the brightness obtained by simultaneously increasing the brightness of all three color channels. This represents the amount of positive brightness added by all three color channels combined. This represents the maximum color channel value before the brightness was increased; This represents the minimum color channel value before increasing the brightness.

[0033] As can be seen from the above relationship, under the condition of superposition of near-neutral brightness components, brightness increases, but the maximum difference between the original three color channels does not increase, resulting in a decrease in saturation. This change does not necessarily occur in all bright areas and still needs to be judged in conjunction with the actual calculation results within the candidate region.

[0034] In the first embodiment, to reduce the impact of local abnormal pixels, isolated patterns, and edge pixels in the highlight candidate region on the statistical results, the median is used as the luminance and saturation statistics for the low-brightness pixel group and the high-brightness pixel group. The luminance median of the low-brightness pixel group and the high-brightness pixel group are calculated separately. The luminance level difference is obtained by subtracting the luminance median of the low-brightness pixel group from the luminance median of the high-brightness pixel group, and this luminance level difference characterizes the degree of brightness increase of the high-brightness pixel group relative to the low-brightness pixel group. Its mathematical expression is as follows: ; in, Indicates the first The brightness level difference of each bright candidate region; Indicates the first The median brightness of the high-brightness pixel group in the candidate brightness region; Indicates the first The median brightness of the low-brightness pixel group in the bright candidate region; This indicates the sequence number of the candidate region to be highlighted.

[0035] Brightness level difference is used to represent the degree of brightness separation between high-brightness pixel groups and low-brightness pixel groups within the same bright candidate region. The larger the brightness level difference, the more obvious the brightness change between the two groups of pixels; when the brightness level difference is close to 0, it indicates that there is no obvious hierarchical relationship from lower brightness to higher brightness within the candidate region.

[0036] In the first embodiment, the median saturation of the low-brightness pixel group and the high-brightness pixel group are calculated separately. The saturation decrease is obtained by subtracting the median saturation of the high-brightness pixel group from the median saturation of the low-brightness pixel group, and this saturation decrease represents the degree of saturation reduction of the high-brightness pixel group relative to the low-brightness pixel group. The interquartile range (ICM) of the saturation data in the local reference ring corresponding to the bright candidate region is calculated, and this ICM represents the degree of saturation fluctuation in the local reference ring. Its mathematical expression is as follows: ; in, Indicates the first The saturation decrease of each bright candidate region from low-brightness pixel group to high-brightness pixel group; Indicates the first The median saturation of the low-brightness pixel group in the bright candidate region; Indicates the first The median saturation of the high-brightness pixel group in the candidate region of the highlight.

[0037] The saturation decrease is used to represent the change in color saturation between high-brightness pixel groups and low-brightness pixel groups. When the saturation decrease is greater than 0, it means that the saturation of the high-brightness pixel group is lower than that of the low-brightness pixel group; when the saturation decrease is close to 0, it means that the saturation between the two brightness levels is basically the same; when the saturation decrease is less than 0, there is no relationship of brightness increase accompanied by saturation decrease.

[0038] For the first The interquartile range of saturation data in the local reference ring corresponding to each bright candidate region is calculated and used as the benchmark for local saturation fluctuation on the surfaces of adjacent artifacts. When the brightness level difference is greater than 0, the saturation decrease is greater than 0, and the saturation decrease is greater than the saturation interquartile range of the corresponding local reference ring, the bright candidate region is determined to meet the condition of brightness increase accompanied by saturation decrease.

[0039] The difference in saturation median between the low-brightness pixel group and the high-brightness pixel group in the bright candidate region represents the overall change in saturation position between the two brightness levels. The saturation interquartile range of the local reference ring represents the discrete range of the middle 50% saturation data of adjacent artifact surfaces. When the aforementioned difference in saturation median is greater than the saturation interquartile range of the local reference ring, it indicates that the saturation change in the candidate region with increasing brightness level exceeds the main saturation fluctuation range of adjacent surfaces.

[0040] For surfaces of collectibles that are nearly colorless and lack effective saturation differences between different brightness levels, this step may not be able to determine the specular highlight area solely based on saturation changes. Therefore, such surfaces are not the primary applicable objects for this embodiment to distinguish highlights based solely on the brightness-saturation relationship.

[0041] At this point, step S2 is complete, and the output shows the candidate bright areas that meet the condition of increased brightness accompanied by decreased saturation, as well as the corresponding low-brightness pixel groups, high-brightness pixel groups, brightness level differences, and saturation decrease amounts.

[0042] S3: Perform gradient attenuation verification based on layers with the same brightness and determine the specular highlight area; It should be noted that when there is strong specular reflection on the surfaces of ceramic glazes, polished jade, lacquerware, and metal artifacts, some pixel values ​​in the bright areas may approach the upper limit of the display range. This compresses the local differences in brightness caused by scratches, fine lines, glaze textures, or surface undulations. Therefore, the gradient response of higher brightness areas may be lower than that of lower brightness areas in the same region. Although real light-colored patterns may also have high brightness, their boundaries, lines, or local textures may still maintain corresponding gradient changes.

[0043] Specifically, the Sobel operator is used to calculate the gradient magnitude of each pixel in the brightness component of the collection image.

[0044] In the first embodiment, the median gradient magnitude is used as the gradient magnitude statistic for the low-brightness pixel group and the high-brightness pixel group. For the i-th high-brightness candidate region that satisfies the conditions described in step S2, the gradient magnitudes of the corresponding pixels in the low-brightness pixel group and the high-brightness pixel group are read respectively, and the median gradient magnitude of each group is calculated. The gradient descent amount is obtained by subtracting the median gradient magnitude of the high-brightness pixel group from the median gradient magnitude of the low-brightness pixel group. ; in, Indicates the first The gradient descent amount from low-brightness pixel group to high-brightness pixel group for each bright candidate region; Indicates the first The median gradient magnitude of the low-brightness pixel group in the bright candidate region; Indicates the first The median gradient magnitude of the high-brightness pixel group in the candidate region of the highlight.

[0045] Gradient descent is used to represent the degree of reduction in local grayscale change and edge response of high-brightness pixel groups relative to low-brightness pixel groups within the same highlight candidate region. A gradient descent greater than 0 indicates that the gradient response decreases as the brightness level increases; a gradient descent close to 0 indicates that the local texture response of the two brightness levels is basically the same; a gradient descent less than 0 indicates that the gradient response of the high-brightness pixel group is not weakened or even enhanced.

[0046] The gradient interquartile range is calculated based on the gradient magnitude of each pixel in the local reference ring corresponding to the candidate highlight region. When the gradient descent is greater than 0 and greater than the gradient interquartile range, the candidate highlight region is determined to meet the gradient decay check condition, and the candidate highlight region that simultaneously meets the conditions described in step S2 and the gradient decay check condition is determined as a specular highlight region. Candidate highlight regions that do not meet either condition retain their original image data and are not included in subsequent specular highlight correction.

[0047] Using the median gradient instead of a single maximum gradient can reduce the impact of candidate region edges, isolated scratches, or random noise on the texture state; using the gradient interquartile range of the corresponding local reference ring as a comparison benchmark allows the gradient descent amount to be judged relative to the actual texture fluctuation of the local surface of the current collection, rather than using a fixed gradient threshold uniformly set for all collections.

[0048] This embodiment is applicable to images of artifacts where detectable brightness level differences and corresponding saturation changes are still present within the candidate bright area. When specular highlights cause all pixels within the candidate area to reach the same channel saturation value, there are no effective brightness or saturation differences within the candidate area for brightness level determination in step S2. Therefore, such fully saturated areas are not the primary applicable objects for specular highlight differentiation based on the brightness-saturation relationship in this embodiment. For pixels that have already undergone channel truncation, this embodiment does not describe subsequent processing as restoring color or texture data lost in the original acquired image.

[0049] In the second embodiment, the determination of the bright candidate region, the acquisition of the local reference ring, and the division of the low-brightness pixel group and the high-brightness pixel group are the same as in the first embodiment. The difference is that the arithmetic mean is used as the brightness statistic, saturation statistic, and gradient magnitude statistic of the low-brightness pixel group and the high-brightness pixel group, and the standard deviation is used as the specific characterization of the saturation fluctuation degree and gradient fluctuation degree of the local reference ring.

[0050] Specifically, the normalized average brightness of the low-brightness pixel group and the high-brightness pixel group are calculated separately. The normalized average brightness of the low-brightness pixel group is subtracted from the normalized average brightness of the high-brightness pixel group to obtain the brightness increase in the second embodiment. The brightness increase is used to characterize the degree of brightness increase of the high-brightness pixel group relative to the low-brightness pixel group.

[0051] The normalized average saturation of the low-brightness pixel group and the high-brightness pixel group are calculated separately. The normalized average saturation of the low-brightness pixel group is subtracted from the normalized average saturation of the high-brightness pixel group to obtain the saturation reduction amount in the second embodiment. The saturation reduction amount is used to characterize the degree of saturation reduction of the high-brightness pixel group relative to the low-brightness pixel group.

[0052] The standard deviation of the normalized saturation data in the local reference ring corresponding to the bright candidate region is calculated, and the saturation standard deviation is used to characterize the degree of saturation fluctuation in the local reference ring. When both the increase in brightness and the decrease in saturation are greater than 0, and the decrease in saturation is greater than the saturation standard deviation of the corresponding local reference ring, the bright candidate region is determined to meet the condition of increased brightness accompanied by decreased saturation.

[0053] For the high-brightness candidate regions that meet the condition of increased brightness accompanied by decreased saturation, the average gradient magnitude of the corresponding pixels in the low-brightness pixel group and the high-brightness pixel group is calculated respectively. The average gradient magnitude of the low-brightness pixel group is subtracted from the average gradient magnitude of the high-brightness pixel group to obtain the gradient reduction amount in the second embodiment. The gradient reduction amount is used to characterize the degree of gradient reduction of the high-brightness pixel group relative to the low-brightness pixel group.

[0054] The standard deviation of the gradient magnitude of each pixel in the local reference ring corresponding to the candidate bright area is calculated, and the standard deviation of the gradient magnitude is used to characterize the gradient fluctuation of the local reference ring. When the gradient reduction is greater than 0 and greater than the standard deviation of the gradient magnitude of the corresponding local reference ring, the candidate bright area is determined as a specular highlight area; candidate bright areas that do not meet the above conditions retain the original image data and are not included in the subsequent specular highlight correction.

[0055] The second implementation uses the same logic as the first implementation. First, the degree of brightness increase and saturation decrease are determined based on the low-brightness pixel group and the high-brightness pixel group in the same bright candidate region. The degree of saturation decrease is then compared with the degree of saturation fluctuation of the corresponding local reference ring. For bright candidate regions that meet the condition of brightness increase accompanied by saturation decrease, the degree of gradient decrease is then determined based on the same low-brightness pixel group and the high-brightness pixel group. The degree of gradient decrease is then compared with the degree of gradient fluctuation of the corresponding local reference ring.

[0056] The difference between the first and second implementation methods lies in the different statistical representation methods used for the overall level of the pixel group and the dispersion of the local reference ring. The first implementation uses the median to represent the overall level of the pixel group and the interquartile range to represent the data fluctuation of the local reference ring; the second implementation uses the arithmetic mean to represent the overall level of the pixel group and the standard deviation to represent the data fluctuation of the local reference ring. Both implementation methods are used to determine whether the color and texture changes generated when transitioning from a low brightness level to a high brightness level exceed the normal local fluctuations of the adjacent surfaces of the current artifact.

[0057] At this point, step S3 is completed, and the specular highlight area, as well as the corresponding low-brightness pixel group, high-brightness pixel group, brightness increase degree, saturation decrease degree, and local reference ring are output; wherein, in the first embodiment, the brightness increase degree and saturation decrease degree are characterized by brightness level difference and saturation decrease amount, respectively; in the second embodiment, the brightness increase degree and saturation decrease degree are characterized by brightness increase amount and saturation decrease amount, respectively.

[0058] S4: Correct the specular highlight area based on the degree of brightness increase and saturation decrease and output a visually enhanced image; It should be noted that when the surface of an artifact is affected by specular reflection, the degree of impact from highlights varies across different locations within the same bright area. Pixels with higher brightness are relatively more affected by highlights, while pixels in the lower brightness areas of similar regions may retain more of their original color and texture information. Furthermore, the candidate bright areas may also contain light-colored patterns, light-colored glazes, or light-colored materials from the artifact itself. Therefore, this step only performs visual enhancement processing on the specular highlight areas identified after brightness and saturation change judgment and gradient attenuation verification. The corresponding local correction degree is determined based on the degree of brightness increase and saturation decrease obtained during the aforementioned brightness layering process, ensuring that the color change features obtained during specular highlight area identification can continue to be used for subsequent brightness and saturation correction.

[0059] In the first embodiment, for the first For each specular highlight area, the brightness level difference obtained in step S2 is used as the brightness reduction reference amount, and the saturation reduction amount obtained in step S2 is used as the saturation compensation reference amount. The brightness level difference is a specific representation of the degree of brightness increase in the first embodiment, and the saturation reduction amount is a specific representation of the degree of saturation decrease in the first embodiment.

[0060] In the first embodiment, pixels whose normalized brightness is higher than the median brightness of the corresponding low-brightness pixel group are identified as pixels to be corrected, and the median brightness of the low-brightness pixel group is used as the starting brightness reference for the correction ratio. The correction ratio is calculated based on the brightness position of each pixel in the specular highlight area relative to the median brightness. The correction ratio for the pixels to be corrected is greater than 0, while the correction ratio for pixels whose normalized brightness is not higher than the median brightness is 0. No actual correction is performed on the latter, and the corresponding original color data is retained.

[0061] In the second embodiment, for the first For each specular highlight area, the brightness increase obtained in the second embodiment is used as the brightness reduction reference amount, and the saturation reduction obtained in the second embodiment is used as the saturation compensation reference amount. The brightness increase amount is a specific representation of the degree of brightness increase in the second embodiment, and the saturation reduction amount is a specific representation of the degree of saturation reduction in the second embodiment.

[0062] In the second embodiment, pixels whose normalized brightness is higher than the average brightness of the corresponding low-brightness pixel group are identified as pixels to be corrected, and the average brightness of the low-brightness pixel group is used as the starting brightness reference for the correction ratio. For the pixel to be corrected, the brightness difference is obtained by subtracting the average brightness of the corresponding low-brightness pixel group from the normalized brightness of the pixel to be corrected. The correction denominator is obtained by subtracting the average brightness of the low-brightness pixel group from 1 and adding a preset positive number to make the correction denominator greater than 0. The brightness difference is divided by the correction denominator, and the resulting ratio is limited to the range of 0 to 1 to obtain the correction ratio in the second embodiment.

[0063] In the second embodiment, the product of the correction ratio and the brightness reduction reference amount is used as the brightness reduction amount. The brightness reduction amount is subtracted from the normalized brightness of the pixel to be corrected before correction, and the resulting value is limited to the range of 0 to 1 to obtain the normalized brightness of the pixel to be corrected after correction.

[0064] For pixels to be corrected in the second embodiment whose pre-correction normalized saturation is greater than 0, the product of the correction ratio and the saturation compensation reference amount is added to the pre-correction normalized saturation to obtain the initial saturation correction value; the larger value between the pre-correction normalized saturation and the average saturation of the corresponding low-brightness pixel group is taken as the upper limit of saturation, and the smaller value between the initial saturation correction value and the upper limit of saturation is taken as the post-correction normalized saturation, while keeping the hue component unchanged.

[0065] In the second embodiment, when the normalized saturation of the pixel to be corrected before correction is equal to 0, the normalized saturation after correction is set to 0, and the hue component remains unchanged, only the brightness is corrected.

[0066] Both the first and second embodiments determine the brightness reduction benchmark based on the degree of brightness increase, the saturation compensation benchmark based on the degree of saturation decrease, and the correction ratio based on the position of the pixel to be corrected relative to the corresponding brightness statistics of the low-brightness pixel group. The difference is that the first embodiment uses the median of the low-brightness pixel group as the corresponding statistical benchmark, while the second embodiment uses the arithmetic mean of the low-brightness pixel group as the corresponding statistical benchmark.

[0067] The following explanation, using the first embodiment as an example, details the brightness reduction reference amount, saturation compensation reference amount, correction ratio, and the specific correction relationship between brightness and saturation. For the i-th specular highlight region, the brightness level difference obtained in step S2 is used as the brightness reduction reference amount, and the saturation decrease obtained in step S2 is used as the saturation compensation reference amount. ; ; in, Indicates the first The baseline amount of brightness reduction for each specular highlight area; This indicates that the result obtained in step S2 is the first... The brightness levels of individual specular highlight areas; Indicates the first Saturation compensation baseline for each specular highlight area; This indicates that the result obtained in step S2 is the first... The amount of saturation reduction in the specular highlight area.

[0068] The brightness level difference represents the increase in brightness of the high-brightness pixel group relative to the low-brightness pixel group, and is used to determine the brightness benchmark that needs to be reduced; the saturation reduction represents the decrease in saturation of the high-brightness pixel group relative to the low-brightness pixel group, and is used to determine the saturation benchmark that needs to be compensated.

[0069] To avoid applying the same correction magnitude to all pixels within the specular highlight area, a correction ratio between 0 and 1 is calculated for each pixel within the specular highlight area based on its brightness position: ; in, Represents pixels in the specular highlight area The correction ratio; Represents pixels Normalized brightness; This indicates that the result obtained in step S2 is the first... The median brightness of the low-brightness pixel group in the specular highlight area; This represents a preset positive number used to make the corrected denominator greater than 0. In this embodiment, it is taken as... ; This indicates a truncation operation that restricts the calculation result to the range of 0 to 1.

[0070] The median brightness of the low-brightness pixel group is used as the starting point for the correction ratio. When the pixel brightness is not higher than this median brightness, the correction ratio is limited to 0, and no brightness reduction is applied. As the pixel brightness increases above this median brightness, the correction ratio increases accordingly, thereby allowing the higher brightness areas within the specular highlight area to receive a larger correction range, while preserving the existing brightness differences in the lower brightness areas.

[0071] The brightness of each pixel to be corrected within the specular highlight area is corrected according to the following formula: ; in, Represents pixels Corrected normalized brightness; Represents pixels Normalized brightness before correction; Represents pixels The correction ratio; Indicates the first The baseline amount of brightness reduction for each specular highlight area; This indicates a truncation operation that restricts the calculation result to the range of 0 to 1.

[0072] After completing the brightness correction, for pixels whose saturation was greater than 0 before correction, the initial saturation correction value is obtained according to the correction ratio and the saturation compensation baseline, and the upper limit of saturation compensation is limited by the median saturation of the low-brightness pixel group. when hour, , ; when hour, , Only brightness correction is performed.

[0073] in, Represents pixels Corrected normalized saturation; Represents pixels Normalized saturation before correction; Represents pixels Corrected color tone; Represents pixels The original color tone; Represents pixels The correction ratio; Indicates the first Saturation compensation baseline for each specular highlight area; This indicates that the result obtained in step S2 is the first... The median saturation of low-brightness pixel groups in the specular highlight area; This indicates that the smaller value is selected from the values ​​being compared. This indicates that the larger value is selected from the values ​​being compared.

[0074] When the saturation level before pixel correction is not higher than the median saturation level of the low-brightness pixel group, the result after saturation compensation will not exceed that median. When the saturation level before pixel correction is already higher than the median, the corrected saturation level will remain unchanged, thus avoiding further increasing the saturation of pixels that originally had high saturation. For pixels with a saturation level of 0 before correction, only the brightness is corrected while the saturation level remains at 0; no new color information is inferred from pixels with no effective color difference.

[0075] The luminance and saturation components have been normalized to the range of 0 to 1 in step S1. The corrected luminance is limited to the range of 0 to 1 through a truncation operation; the corrected saturation is maintained within the range of 0 to 1 through the aforementioned larger and smaller value operations, thereby satisfying the subsequent... Input requirements for color space inverse conversion.

[0076] After completing the local correction of the pixels to be corrected in the specular highlight area, the corrected... Data inverse conversion Data. For pixels to be corrected, the corrected and converted data is used. Color data; for pixels in non-specular highlight areas and pixels in specular highlight areas that were not identified as pixels to be corrected, retain the corresponding original data. Color data is used to obtain visually enhanced images of digital museum collections based on feature fusion.

[0077] It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this invention, and these modifications and improvements are all within the scope of protection of this invention.

Claims

1. A method for visually enhancing digital museum collections based on feature fusion, characterized in that: include: Obtain the subject-cropped Collection images and conversion In a color space, candidate bright areas are determined based on luminance components, and a local reference ring around the candidate bright areas is obtained. For the bright candidate region with local reference ring, divide it into low brightness pixel group and high brightness pixel group according to brightness. Based on the brightness statistics and saturation statistics of the two groups of pixels, determine the degree of brightness increase and saturation decrease of the high brightness pixel group relative to the low brightness pixel group. Combine the degree of saturation fluctuation of local reference ring to screen the bright candidate region that meets the condition of brightness increase accompanied by saturation decrease. For a bright candidate region that meets the condition of increased brightness accompanied by decreased saturation, the pixel gradient magnitude is calculated based on the brightness component, and the gradient reduction degree of the high brightness pixel group relative to the low brightness pixel group is determined based on the gradient magnitude statistics of the two groups of pixels. The specular highlight region is determined by combining the gradient fluctuation degree of the corresponding local reference ring. The degree of local correction is determined based on the increase in brightness and decrease in saturation corresponding to the specular highlight area. The brightness and saturation components of the specular highlight area are then corrected according to this degree of local correction. Data conversion Data, and retain the original pixels without local correction. The data is used to obtain visually enhanced images.

2. The method for visual enhancement of digital museum collections based on feature fusion according to claim 1, characterized in that, When the increase in brightness is positive and the decrease in saturation exceeds the saturation fluctuation of the corresponding local reference ring, the bright candidate region is determined to meet the condition of increased brightness accompanied by decreased saturation. When the gradient reduction is positive and exceeds the gradient fluctuation of the corresponding local reference ring, the bright candidate region that satisfies the condition of increased brightness accompanied by decreased saturation is determined as the specular highlight region.

3. The method for visual enhancement of digital museum collections based on feature fusion according to claim 2, characterized in that, The image of the artifact is an eight-bit, three-channel image of the artifact acquired under white light illumination and with the main subject cropped; the image of the artifact is then converted to... After color space is applied, the luminance and saturation components are normalized to 0 to 1; The process of determining the candidate region for brightness includes: sorting the brightness components, taking the upper quartile of the brightness components as the candidate region for brightness, dividing adjacent pixels with brightness not lower than the candidate region for brightness into connected regions according to the eight-neighbor connectivity relationship, and determining connected regions with no less than four pixels as candidate regions for brightness.

4. The method for visual enhancement of digital museum collections based on feature fusion according to claim 3, characterized in that, Morphological dilation is performed on each highlight candidate region to extend the boundary of the highlight candidate region outward by 5 pixels. The newly added adjacent pixels that do not belong to other highlight candidate regions are used as the local reference ring of the highlight candidate region. For candidate bright areas where local reference ring pixels cannot be obtained, retain the corresponding original image data and do not proceed to subsequent specular highlight confirmation.

5. The method for visual enhancement of digital museum collections based on feature fusion according to claim 3, characterized in that, The pixels in each bright candidate region are sorted from smallest to largest according to their normalized brightness values. The pixels in the first half of the sorted region are identified as the low brightness pixel group, and the pixels in the second half of the sorted region are identified as the high brightness pixel group. When the bright candidate region contains an odd number of pixels, the pixels in the middle sorted position are assigned to the low brightness pixel group. The brightness level difference is obtained by subtracting the brightness median of the low brightness pixel group from the brightness median of the high brightness pixel group, and the brightness level difference is used as the degree of brightness increase. The saturation decrease is obtained by subtracting the saturation median of the high-brightness pixel group from the saturation median of the low-brightness pixel group, and the saturation decrease is used as the degree of saturation reduction. The saturation interquartile range is calculated based on the saturation data in the local reference ring corresponding to the bright candidate region, and the saturation interquartile range is used as the degree of saturation fluctuation.

6. The method for visual enhancement of digital museum collections based on feature fusion according to claim 5, characterized in that, The Sobel operator is used to calculate the gradient magnitude of each pixel in the brightness component of the collection image; For a bright candidate region that satisfies the condition of increased brightness accompanied by decreased saturation, the median gradient magnitude of the corresponding pixels in the low-brightness pixel group and the high-brightness pixel group are calculated respectively. The gradient decrease amount is obtained by subtracting the median gradient magnitude of the high-brightness pixel group from the median gradient magnitude of the low-brightness pixel group, and the gradient decrease amount is used as the degree of gradient reduction. The gradient interquartile range is calculated based on the gradient magnitude of each pixel in the local reference ring corresponding to the bright candidate region, and the gradient interquartile range is used as the gradient fluctuation degree.

7. The method for visual enhancement of digital museum collections based on feature fusion according to claim 6, characterized in that, The degree of local correction is determined by the brightness reduction reference amount, the saturation compensation reference amount, and the correction ratio of the pixel to be corrected; For each specular highlight area, the difference in brightness level is used as the brightness reduction reference amount, the decrease in saturation is used as the saturation compensation reference amount, and pixels whose normalized brightness is higher than the median brightness of the corresponding low brightness pixel group are identified as pixels to be corrected. For the pixel to be corrected, the brightness difference is obtained by subtracting the median brightness of the corresponding low-brightness pixel group from the normalized brightness of the pixel to be corrected. The correction denominator is obtained by subtracting the median brightness of the low-brightness pixel group from 1 and adding a preset positive number to make the correction denominator greater than 0. The brightness difference is divided by the correction denominator, and the resulting ratio is limited to the range of 0 to 1 to obtain the correction ratio.

8. The method for visual enhancement of digital museum collections based on feature fusion according to claim 7, characterized in that, The product of the correction ratio and the brightness reduction reference amount is used as the brightness reduction amount. The brightness reduction amount is subtracted from the normalized brightness of the pixel to be corrected before correction, and the resulting value is limited to the range of 0 to 1 to obtain the normalized brightness of the pixel to be corrected after correction.

9. The method for visual enhancement of digital museum collections based on feature fusion according to claim 8, characterized in that, When the normalized saturation of the pixel to be corrected is greater than 0, the product of the correction ratio and the saturation compensation reference amount is added to the normalized saturation before correction to obtain the initial saturation correction value. The larger of the normalized saturation before correction and the median saturation of the corresponding low-brightness pixel group is taken as the upper limit of saturation, and the smaller of the initial saturation correction value and the upper limit of saturation is taken as the normalized saturation after correction, while keeping the hue component unchanged. When the normalized saturation of the pixel to be corrected is equal to 0 before correction, the normalized saturation after correction is set to 0, and the hue component remains unchanged, only the brightness is corrected.

Images