Intelligent color homogenizing method based on online remote sensing image service

Abstract

The invention discloses an intelligent color uniformizing method based on online remote sensing image service, and the method comprises the steps: determining the positions and number of reference image tiles hierarchically according to the resolution and latitude and longitude coordinates of an image to be subjected to color uniformizing; calculating a mean value and a variance of each layer of tiles, and generating a cross-scale variable coefficient distribution diagram; performing multi-scale fusion on the reference image based on the statistical characteristics; and carrying out color homogenization on the image to be subjected to color homogenization by taking a fusion result as a reference, outputting a result and evaluating quality. According to the method, image analysis and machine learning technologies are applied, multi-scale automatic color uniformizing is realized, a color uniformizing reference can be quickly positioned, a visual quality inspection report can be output, GPU acceleration and embedded deployment are supported, scenes such as satellite real-time processing and ground station batch color uniformizing are met, and the automation degree and reliability of remote sensing data color uniformizing are improved.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image color balancing technology, and in particular to an intelligent color balancing method based on online remote sensing image services. Background Technology

[0002] During the acquisition of remote sensing images, inconsistencies in color and brightness often occur between multiple images of the same area or between adjacent images due to various factors such as differences in sensor performance, variations in lighting conditions, atmospheric scattering, and different shooting times. These inconsistencies manifest as color casts and uneven brightness. Such problems not only severely affect the visual quality of the images but also reduce the accuracy of subsequent analysis. For example, in applications such as land cover classification and change detection, inconsistencies in color and brightness can lead to misjudgments or missed detections. Therefore, color homogenization processing, as a crucial step in eliminating these inconsistencies and achieving color and brightness uniformity across multiple images, is of great significance in remote sensing image processing.

[0003] With the continuous advancement of algorithm technology, color balancing methods are gradually developing towards automation and intelligence. Currently, the mainstream algorithms fall into the following categories: Reference image-based color balancing: A color standard image is selected as a reference. Regression models (such as multinomial fitting) are used to adjust the color and brightness of other images, ensuring visual consistency between the target image and the reference image. Global optimization methods: Color relationships are established using overlapping regions or feature point matching between images (such as the SIFT algorithm). Overall color consistency is optimized through least squares methods or gradient domain fusion, suitable for stitching and color balancing of large-scale images. Deep learning methods: Deep learning models such as Convolutional Neural Networks (CNNs) or Generative Adversarial Networks (GANs) are used to learn color mapping relationships, achieving end-to-end color balancing processing. For example, the CycleGAN model can be used for color balancing tasks across sensor images, exhibiting strong adaptability.

[0004] Existing methods suffer from several drawbacks: First, difficulty in selecting reference images: The quality of reference images directly affects the consistency of color balancing results, and in large-area remote sensing image color balancing, it is often difficult to find suitable reference images. Second, low processing efficiency: Color balancing tasks at the provincial or global scale require high-performance computing resources (such as GPU clusters). Traditional serial processing methods are extremely inefficient and cannot meet the demands of large-scale data processing. Third, insufficient engineering applications: Most existing color balancing algorithms remain at the level of mathematical models and algorithm research, with limited research on reference image selection methods. There is a lack of solutions that can be directly applied to engineering practice, hindering promotion and implementation.

[0005] For example, invention application No. 202311385513.4 discloses a method, apparatus, device, and storage medium for color balancing of remote sensing images. This application's solution solves the technical problem of low operating efficiency in the prior art when color balancing remote sensing images is performed manually or using professional color balancing software. However, its solution also has the following drawbacks: it relies on a single reference image, which can easily affect the consistency of color balancing due to fluctuations in the quality of the reference image itself in large-area processing; it also does not consider the differences in local features at multiple scales, limiting its applicability to color balancing of remote sensing images over large areas; furthermore, its downsampling process may lose detailed information, resulting in blurred image textures after color balancing, making it difficult to meet the needs of high-precision surveying and mapping scenarios.

[0006] Therefore, there is a need for an intelligent color balancing method and system based on online remote sensing image services. Providing efficient and reliable intelligent color balancing means for common remote sensing satellite data is of great practical significance. Summary of the Invention

[0007] To address the aforementioned problems, the present invention aims to provide an intelligent color balancing method based on online remote sensing image services, which solves the problems of difficulty in selecting reference images, low processing efficiency, and lack of solutions that can be directly applied to engineering practice in existing color balancing methods.

[0008] This invention provides an intelligent color balancing method based on online remote sensing image services.

[0009] First aspect: An intelligent color balancing method based on online remote sensing image services, including the following steps:

[0010] S1. Based on the resolution and latitude and longitude coordinates of the image to be uniformly colored, calculate the location and quantity of reference image tile data in layers.

[0011] S2. Calculate the mean and variance of the reference image tile data at each level;

[0012] S3. Generate a cross-scale coefficient of variation distribution map;

[0013] S4. Perform multi-scale fusion of the reference image based on the mean, variance, and coefficient of variation of the reference image;

[0014] S5. Perform image color balancing processing on the image to be sized based on the reference image after multi-scale fusion.

[0015] S6. Output the color balancing results and perform a quality assessment.

[0016] In one embodiment of the present invention, the calculation of the location and quantity of reference image tile data in step S1 includes the following steps:

[0017] S11. Convert the WGS84 latitude and longitude of the area of ​​the image to be uniformly colored into Web Mercator projection coordinates.

[0018] S12. Calculate the best matching scaling level based on the target resolution;

[0019] S13. Combine the scaling levels to convert Mercator projection coordinates into pixel coordinates, and obtain the position and number of tile data in layers.

[0020] In one embodiment of the present invention, the step of calculating the mean and variance of the reference image tile data at each level in S2 includes the following steps:

[0021] S21. Calculate the independent mean μ and variance σ² of each tile at each level in parallel.

[0022] S22. Synthesize the global mean of tiles at each level by weighted averaging. and global variance ;

[0023] In one embodiment of the present invention, generating a cross-scale coefficient of variation distribution map in step S3 includes the following steps:

[0024] S31. Calculate the independent mean μ and variance σ² for each tile at each level;

[0025] S32. Calculate the coefficient of variation (CV) for each tile based on its independent mean μ and variance σ².

[0026] S23. Summarize the coefficient of variation (CV) statistics for each level of tile and generate a cross-scale coefficient of variation distribution map.

[0027] According to claim 4, the intelligent color balancing method based on online remote sensing image services is characterized in that the coefficient of variation (CV) is expressed by the formula:

[0028]

[0029] Where σ is the standard deviation and μ is the mean.

[0030] In one embodiment of the present invention, multi-scale fusion is performed in S4 by waveband.

[0031] In one embodiment of the present invention, image color balancing processing of the image to be sizing is performed in step S5 using a Wallis filter.

[0032] The mean and variance of the multi-scale fused reference image are used as reference values, and the color uniformity intensity is controlled by adjusting the brightness coefficient and contrast coefficient.

[0033] In one embodiment of the present invention, the color uniformity evaluation quality assessment indicators include color difference analysis, radiation consistency, structural similarity, and local contrast retention.

[0034] Second aspect: An electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, performs the steps of the method provided in the first aspect.

[0035] Third aspect: A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method provided in the first aspect.

[0036] The beneficial effects of this invention are:

[0037] 1. This invention provides an intelligent color balancing tool based on online remote sensing services. Utilizing advanced image analysis and machine learning technologies, it can automatically perform multi-scale color balancing processing. By analyzing the mean, variance, and coefficient of variation (CV) of reference image data at each layer, the system can quickly locate the color balancing reference image and output a visual quality inspection report. It also supports GPU acceleration and embedded deployment, meeting the needs of various scenarios such as real-time satellite processing and automated batch color balancing at ground stations, significantly improving the automation and reliability of intelligent color balancing of remote sensing data.

[0038] 2. This invention utilizes online remote sensing image services to acquire reference tile data. Combined with hierarchical mean, variance, and coefficient of variation analysis, it achieves intelligent selection of reference images, avoiding the shortcomings of traditional methods that rely on manual selection. Simultaneously, through local adaptive adjustment of Wallis filtering and a multi-scale fusion strategy, it effectively eliminates color inconsistencies caused by sensor differences and illumination variations while ensuring local details. This is particularly suitable for color balancing of large-scale images, significantly improving processing efficiency and result consistency.

[0039] 3. This invention employs tiled parallel processing and hierarchical statistical analysis methods. Parallel computing accelerates the calculation of statistical quantities such as mean and variance, solving the efficiency bottleneck of traditional serial processing in ultra-large image processing. Simultaneously, a cross-scale coefficient of variation distribution map is used to quantify local variation features at different resolutions, providing data support for reference image fusion and filter parameter optimization. Furthermore, it offers dual strategies—fast mode and fine mode—to meet the accuracy requirements of automated rapid mapping and surveying-grade products, respectively, achieving a balance between efficiency and accuracy. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating the principle of the intelligent color-balancing method of the present invention;

[0041] Figure 2 This is a flowchart illustrating the intelligent color-balancing method of the present invention;

[0042] Figure 3This is a comparison diagram showing the application effect of the intelligent color-balancing method of the present invention. Figure 4 This is a comparison diagram showing another application effect of the intelligent color-balancing method of the present invention;

[0043] Figure 5 This is a schematic diagram of the electronic device structure of the present invention. Detailed Implementation

[0044] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar symbols denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0045] Existing methods for color balancing remote sensing images suffer from several drawbacks: First, selecting a reference image is difficult, as its quality affects the consistency of the color balancing effect, and it is hard to find a suitable reference image when processing large areas. Second, processing efficiency is low; color balancing of provincial or global scale images requires high-performance computing resources, and traditional serial processing methods cannot meet the needs of large-scale data processing. Third, there is insufficient engineering application; existing color balancing algorithms are mostly at the research level, with little research on reference image selection methods, and a lack of solutions that can be directly applied to engineering practice, making it difficult to promote and implement.

[0046] To address the above problems, this invention provides an intelligent color balancing method based on online remote sensing image services. Figure 1 The flowchart illustrates the principle of the intelligent color-balancing method provided by this invention.

[0047] Example:

[0048] This embodiment discloses an intelligent color balancing method based on online remote sensing image services, such as... Figure 2 As shown, the steps include:

[0049] S1. Based on the resolution and latitude and longitude coordinates of the image to be uniformly colored, calculate the location and quantity of reference image tile data in layers.

[0050] First, based on the resolution and latitude / longitude coordinates of the image to be color-matched, the location and number of reference image tiles are determined layer by layer, including:

[0051] Convert the WGS84 latitude and longitude coordinates of the area to be uniformly colored in the image to Web Mercator projection coordinates.

[0052] By unifying coordinate standards, the geographical area of ​​the image to be color-matched is matched with the tile coordinate system of the online remote sensing image service, laying the foundation for subsequent tile positioning.

[0053] Then, determine the scaling level of the layer. Based on the target resolution of the image to be color-matched, refer to the preset correspondence between resolution and scaling level, and calculate the best matching initial scaling level z. The correspondence between target resolution and scaling level z can be found in Table 1.

[0054] Table 1. Reference Table of Relationship between Resolution and Scaling Level z

[0055] Image scaling levels Map height (unit: pixels) Ground scale (resolution) (meters / pixels) Map scale (based on 96 dpi) 1 512 78,271.5170 1:295,829,353.45 2 1,024 39,135.7585 1:147,914,677.6 3 2,048 19,567.8792 1:73,957,338.73 4 4,096 9,783.9396 1:36,978,669.43 5 8,192 4,891.9698 1:18,489,334.72 6 16,384 2,445.9845 1:9,244,667.36 7 32,768 1,222.9925 1:4,622,333.68 8 65,536 611.4962 1:2,311,166.84 9 131,072 305.7481 1:1,155,583.42 10 262,144 152.8741 1:577,791.71 11 524,288 76.4370 1:288,895.85 12 1,048,576 38.2185 1:144,447.93 13 2,097,152 19.1093 1:72,223.96 14 4,194,304 9.5546 1:36,111.98 15 8,388,608 4.7773 1:18,055.99 16 16,777,216 2.3887 1:9,028.00 17 33,554,432 1.1943 1:4,514.00 18 67,108,864 0.5972 1:2,257.00 19 134,217,728 0.2986 1:1,128.50 20 268,435,456 0.1493 1:564.25 21 536,870,912 0.0746 1:282.12 22 1,073,741,824 0.0373 1:141.06 23 2,147,483,648 0.0187 1:70.53

[0056] For example, an image with a resolution of approximately 1 meter per pixel corresponds to a scaling level of z=18.

[0057] Furthermore, hierarchical expansion is performed to achieve multi-scale analysis. Based on the initial scaling level z, it is expanded to adjacent levels (such as z±1, z±2, z±3, etc.), forming a hierarchical set of scaling levels (such as z=15, 16, 17, 18, 19, 20, 21). Different levels correspond to different resolutions (the higher the level, the higher the resolution, and the smaller the tile coverage). Hierarchical processing can take into account both macroscopic and microscopic image features.

[0058] Then, the pixel coordinates for each level are calculated. For each scaling level z, the transformed Web Mercator projection coordinates are mapped to the pixel coordinates at that level. Specifically:

[0059] Based on the pixel coordinate transformation rules of Web Mercator projection, and combined with the resolution parameter of scaling level z, the projected coordinates (meters) are converted into pixel positions (x_pixel, y_pixel).

[0060] Then, the row and column numbers (x, y) of each level of the tile are determined. The tiles in online remote sensing image services are usually 256×256 pixels. The row and column number (x, y) of each tile is obtained by dividing the pixel coordinates by 256 and rounding down.

[0061] For the pixel coordinates (x_pixel, y_pixel) at a certain level z, the calculation formula is:

[0062] The tile row number is x = floor(x_pixel / 256);

[0063] Tile column number y = floor(y_pixel / 256).

[0064] Example: For a region (116.391°E, 39.907°N), after pixel coordinate transformation at z=18 level, the tile row and column numbers are (218880, 104802).

[0065] Then, the number of tiles in each level is calculated. For each level z, the minimum and maximum values ​​of the tile row and column numbers are calculated based on the pixel coordinates corresponding to the four boundaries of the image to be uniformly colored. The number of tiles in a certain level z is then calculated based on the minimum and maximum values.

[0066] Example: If at level z=18, the row number range is 218880 to 218927, and the column number range is 104802 to 104867, then the number of tiles at this level is:

[0067] (218927-218880+1)×(104867-104802+1)=48×66=3168 sheets.

[0068] Organize the range and number of tile rows and columns for each level (e.g., z-3 to z+3) to form a hierarchical list of reference image tile data, providing a foundation for subsequent tile reading and multi-scale analysis.

[0069] Through the above steps, hierarchical tile positioning and quantity calculation based on the parameters of the image to be color-matched can be achieved, which not only ensures the scale matching between the reference image and the image to be processed, but also provides data support for subsequent multi-scale fusion and adaptive color matching.

[0070] S2. Calculate the mean and variance of the reference image tile data at each level.

[0071] For tiles of a fixed size (e.g., 256×256 pixels), the mean μ (reflecting the brightness of the area) and variance σ² (measuring contrast) are calculated independently for each tile, where μ is the arithmetic mean of the pixel values ​​and σ² is the squared mean of the pixel values ​​deviating from the mean.

[0072] For each tile at each level, the mean (μ) and variance (σ²) of each tile are calculated separately for each band:

[0073] The mean (μ) reflects the overall brightness of the tile, and the calculation formula is:

[0074]

[0075] Where N is the total number of valid pixels in the tile (e.g., 256 × 256 = 65536, excluding Nodata values), Let be the grayscale value of the i-th pixel.

[0076] Variance (σ²) reflects the dispersion of pixel values ​​within a tile, and is calculated using the following formula:

[0077]

[0078] For multi-band images such as RGB, the mean and variance are calculated independently for each band, and the statistical results of the red, green, and blue channels are recorded separately.

[0079] The hierarchical global mean (μ_global) and global variance (σ²_global) are synthesized, and a weighted aggregation is performed on all tiles at the same level based on the statistics of individual tiles:

[0080] Furthermore, the weights can be determined based on the overlap ratio between the tile and the area of ​​the image to be color-matched. Tiles with complete coverage have a weight of 1, while those with partial coverage are weighted according to the overlap area ratio (e.g., a 50% overlap results in a weight of 0.5). The overall brightness level of this layer is obtained through a weighted average, resulting in the layer's global mean. )for:

[0081]

[0082] Where M is the total number of tiles at this level, Let be the mean of the k-th tile.

[0083] Based on the principle of variance decomposition, the overall dispersion of the hierarchy is calculated by combining the variance of a single tile and the global mean, and the global variance is synthesized. ):

[0084]

[0085] The formula includes the variance of a single tile. The deviation term between the tile mean and the global mean ensures that the global variance reflects the differences between tiles.

[0086] Furthermore, cross-level statistical results are stored by storing the global mean and variance of each level (e.g., z-3 to z+3) according to the level number, forming a multi-scale statistical data set, which provides a reference for subsequent multi-scale fusion and Wallis filtering color balancing.

[0087] Furthermore, a parallel approach is adopted to accelerate tile statistics, and a global mean is synthesized through weighted averaging. And calculating global variance based on variance decomposition formula Finally, these statistics are used as target parameters for Wallis filtering, and local pixel values ​​are adjusted through sliding window or integral image methods to balance illumination differences while preserving details.

[0088] This method balances computational efficiency and local adaptability through tile processing, and is suitable for scenarios that require consistent lighting, such as remote sensing image stitching and medical image processing. The selection of tile size requires a trade-off between statistical accuracy and computational cost, and the edge areas can be filled with mirror images.

[0089] S3. Generate a cross-scale coefficient of variation distribution map.

[0090] In multi-scale image analysis, the coefficient of variation (CV) of tile data can be used to characterize local variation features at different resolutions through hierarchical calculation. The generation of cross-scale CV distribution maps is based on the statistical characteristics of reference image tiles at each level. Through hierarchical calculation, data integration, and visualization, the local variation features of images at different resolutions can be intuitively displayed.

[0091] The specific method is as follows: First, for each tile in each layer, calculate its mean μ and standard deviation σ.

[0092] The target layer range is determined based on the core scaling level z, and then extended to adjacent layers (e.g., z-3 to z+3) to cover a multi-scale range from low resolution (macroscale) to high resolution (fine scale). For example, if the core layer z=18, the target layers include 15, 16, 17, 18, 19, 20, and 21, for a total of 7 layers.

[0093] By reading the tile data of each level, it is ensured that the tile range of each level is consistent with the area of ​​the image to be color-matched, and invalid values ​​(Nodata) are filtered out.

[0094] The coefficient of variation (CV) is calculated hierarchically for each tile at each level, based on the previously calculated mean. and variance Perform single-tile CV calculation:

[0095] The coefficient of variation (CV) is characterized by the ratio of the standard deviation to the mean, and the formula is:

[0096]

[0097] in, Standard deviation If the mean is... (For example, in a pure black area), CV is set to 0 or Nodata to avoid calculation errors.

[0098] Furthermore, for multi-band images such as RGB, the CV value is calculated independently for each band (e.g., red band CV, green band CV, blue band CV) to reflect the relative dispersion of each channel.

[0099] Furthermore, CV data at low-resolution levels (such as z-3) are upsampled, or CV data at high-resolution levels (such as z+3) are downsampled, so that the data at each level can be compared at the pixel scale. For example, CV data at z=15 level (resolution of approximately 16 meters / pixel) is upsampled to the resolution of z=18 level (approximately 1 meter / pixel), and bilinear interpolation is used to maintain spatial continuity.

[0100] Furthermore, for each level, using tiles as the basic unit, the CV values ​​are mapped to color gradients (e.g., blue, green, and red represent CV from low to high), generating a CV spatial distribution map for that level. For example, areas with CV < 10% are displayed in blue (uniform data), while areas with CV > 30% are displayed in red (complex textures or more noise).

[0101] Furthermore, by using layer overlay technology, CV distribution maps at different levels are integrated into the same coordinate system, highlighting multi-scale features. For example, a high-resolution CV distribution at z=21 is overlaid on a base layer at z=18, clearly displaying local fine variations.

[0102] Through the above steps, the cross-scale coefficient of variation distribution map can clearly present the local variation patterns of images at different resolutions, reflecting both large-scale macro-topographic differences and capturing fine-scale detail fluctuations. It is suitable for scenarios such as remote sensing image quality assessment and adaptive filter parameter optimization. Among them, the hierarchical division needs to balance computational efficiency and feature preservation integrity according to application requirements, providing key support for the selection of reference hierarchical levels and the optimization of filter parameters in intelligent color balancing.

[0103] S4. Perform multi-scale fusion of the reference image based on the mean, variance, and coefficient of variation of the reference image.

[0104] First, the mean and variance of the image data to be uniformly colored are calculated. The calculation of the mean and variance of the image data to be uniformly colored is a key step in the statistical analysis of image color. The mean reflects the central tendency of the image pixel values. By taking the arithmetic mean of all pixel values, the overall brightness level of the entire image can be quantified. The variance measures the degree of dispersion of pixel values ​​around the mean. Its calculation is based on the square mean of the difference between each pixel value and the mean, which can characterize the contrast and color distribution uniformity of the image.

[0105] For multi-band imagery (such as RGB), the mean and variance are calculated independently for each band to evaluate the color characteristics of each channel. This calculation provides a quantitative basis for subsequent color balancing processes (such as histogram matching and color normalization), ensuring the accuracy of color consistency adjustments between different imagery. During the calculation, it is important to exclude invalid pixels (such as Nodata values) to avoid statistical bias.

[0106] Multi-scale fusion combines the statistical characteristics of the mean, variance, and coefficient of variation of reference images at different levels to achieve cross-scale information complementarity, providing a more robust reference benchmark for subsequent color balancing processing.

[0107] Collect statistical results for reference image tiles at each level (e.g., z-3 to z+3), including:

[0108] The mean of a single tile and the entire area ( , ) and variance ( , The coefficient of variation (CV) reflects the brightness level and contrast at different scales; the coefficient of variation (CV) and statistics (such as mean and extreme values) of each level quantify the degree of local variation at different resolutions. The lower the CV value, the more uniform the data, and the more suitable it is as a reference benchmark.

[0109] The fusion weights for each level are determined based on the coefficient of variation, using the following formula:

[0110]

[0111] in, Let k be the weight of the k-th level. Here, is the average coefficient of variation for that level, and n is the total number of levels. For example, if the CV for level z=18 is 15% and the CV for level z=17 is 20%, then level z=18 has a higher weight and its statistical characteristics are retained preferentially.

[0112] Multi-scale mean and variance fusion is performed, and the global means of each level are weighted and summed. The resulting global mean is:

[0113]

[0114] Combining the principle of variance decomposition, and considering both the global variance and mean deviation at each level, the fused global variance is:

[0115]

[0116] The above formula preserves the local contrast information of each level and balances the cross-scale brightness differences through the mean deviation term.

[0117] For tiles corresponding to spatial locations (such as tiles covering the same geographical area in different layers), a sliding window weighted average is used to fuse their mean and variance. The window weights are dynamically adjusted according to the layer resolution, with higher weights assigned to the center of the window for high-resolution layers (such as z+2 and z+3) to ensure that detailed features are preserved.

[0118] The image to be uniformly colored is divided into multiple sub-regions. The average CV value of each sub-region at each level is calculated, and the level with the smallest CV in the sub-region is selected as the main reference scale. For example, densely populated urban areas have lower CV at the high-resolution level (z+2) (more regular texture), while mountainous areas have lower CV at the medium-to-low resolution level (z-1) (more stable terrain trend).

[0119] Furthermore, the scale boundaries of adjacent sub-regions are smoothed using Gaussian blur weighted transition to avoid obvious scale discontinuities after fusion. For example, in the boundary region between z=18 and z=19, the pixel weights transition linearly from z=18 to z=19 to ensure the spatial continuity of statistical features.

[0120] Furthermore, for tiles at the image edge, their neighborhood information is expanded by mirroring before fusion calculation is performed to avoid statistical biases caused by edge effects (such as abrupt changes in the mean value of boundary tiles).

[0121] Furthermore, the cross-level deviation of the global mean and variance after fusion (such as the RMSE of the original statistics of each level) is calculated. If the deviation exceeds the threshold, the weight allocation is readjusted (such as increasing the weight ratio of the lower CV level).

[0122] Through the above process, multi-scale fusion fully utilizes the advantages of different levels of reference images: the stable brightness features of the low CV level ensure global consistency, the detailed statistics of the high-resolution level preserve local texture, and finally form a reference benchmark that combines stability and fineness, laying the foundation for intelligent color balancing of large-scale remote sensing images.

[0123] S5. Perform image color balancing processing on the reference image after multi-scale fusion and the image to be sizing.

[0124] Wallis filtering is a local adaptive color balancing algorithm that balances illumination differences by dynamically adjusting the local mean and variance of an image. Its core idea is to stretch the contrast and normalize the brightness of each pixel neighborhood while preserving edge and texture features.

[0125] The merged global mean ,variance The local tile statistics are stored as a multi-scale reference benchmark. The target mean μ0 and variance σ0² are introduced as ideal reference values ​​and used as target parameters for subsequent Wallis filtering for color uniformity. The filter uses a Gaussian weighted window to calculate local statistics and controls the color uniformity intensity by adjusting parameters A (brightness coefficient) and B (contrast coefficient). When B=1, it completely matches the target variance, and when A=1, it completely matches the target mean. Thus, it preserves image detail features while eliminating uneven illumination and is suitable for color uniformity processing of multi-temporal remote sensing images.

[0126] Furthermore, the weight distribution and scale selection diagrams during the fusion process are output to provide a basis for tracing the uniformity of color quality.

[0127] Furthermore, the Wallis filtering algorithm is used to perform color balancing on the image and output a quality report.

[0128] S6. Output the color balancing results and perform a quality assessment.

[0129] The system ultimately generates an automated color uniformity report: Color uniformity assessment typically employs a combination of subjective and objective methods, with objective indicators including quantitative parameters such as color difference analysis (e.g., ΔE*ab), radiometric consistency (RMSE / PSNR), structural similarity (SSIM), and local contrast retention.

[0130] like Figure 3 and Figure 4 The images show a comparison of the before and after color balancing in Hefei and Lhasa. As can be seen from the images, the unbalanced images exhibit significant color deviations and uneven brightness. For example, some areas in Hefei are too dark, some color blocks in Lhasa are abrupt, and the tonal transitions between different areas are abrupt, severely affecting the overall harmony of the image. After processing with the method of this invention, the color and brightness of the images in each region are significantly improved: the overall brightness of the Hefei image is balanced, and the outlines of ground features are clearly distinguishable; the color block differences in Lhasa are effectively eliminated, the color transitions of natural landscapes are smooth, the image tones are unified, the stitching marks between different areas are basically eliminated, and the visual effect is greatly improved.

[0131] These application results demonstrate that the method of the present invention can effectively solve the problem of color inconsistency in remote sensing images in different regions and scenarios. While maintaining the detailed features of ground features, it significantly improves the overall coordination and visual quality of the images, fully verifying the effectiveness and reliability of the method in practical applications, and providing strong support for the intelligent processing of large-scale remote sensing images.

[0132] The present invention also provides an electronic device, Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 5 As shown, the electronic device may include a processor, a communication interface, a memory, and a communication bus, wherein the processor, communication interface, and memory communicate with each other through the communication bus. The processor can call logical instructions in the memory, for example, to execute the following method: S1, calculate the location and quantity of reference image tile data in layers according to the resolution and latitude and longitude coordinates of the image to be uniformly colored;

[0133] S2. Calculate the mean and variance of the reference image tile data at each level;

[0134] S3. Generate a cross-scale coefficient of variation distribution map;

[0135] S4. Perform multi-scale fusion of the reference image based on the mean, variance, and coefficient of variation of the reference image;

[0136] S5. Perform image color balancing processing on the image to be sized based on the reference image after multi-scale fusion.

[0137] S6. Output the color balancing results and perform a quality assessment.

[0138] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0139] This invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it is implemented to perform the methods provided in the above embodiments, such as including: S1, calculating the location and quantity of reference image tile data in layers according to the resolution and latitude and longitude coordinates of the image to be uniformly colored;

[0140] S2. Calculate the mean and variance of the reference image tile data at each level;

[0141] S3. Generate a cross-scale coefficient of variation distribution map;

[0142] S4. Perform multi-scale fusion of the reference image based on the mean, variance, and coefficient of variation of the reference image;

[0143] S5. Perform image color balancing processing on the image to be sized based on the reference image after multi-scale fusion.

[0144] S6. Output the color balancing results and perform a quality assessment.

[0145] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0146] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent color balancing method based on online remote sensing image services, characterized in that, Including the following steps: S1. Based on the resolution and latitude and longitude coordinates of the image to be uniformly colored, calculate the location and quantity of reference image tile data in layers. S2. Calculate the mean and variance of the reference image tile data at each level; S3. Generate a cross-scale coefficient of variation distribution map; S4. Perform multi-scale fusion of the reference image based on the mean, variance, and coefficient of variation of the reference image; S5. Perform image color balancing processing on the image to be sized based on the reference image after multi-scale fusion. S6. Output the color uniformity result and perform a quality assessment.

2. The intelligent color-balancing method according to claim 1, characterized in that, The calculation of the location and quantity of reference image tile data in S1 includes the following steps: S11. Convert the WGS84 latitude and longitude of the area of ​​the image to be uniformly colored into Web Mercator projection coordinates. S12. Calculate the best scaling level based on the target resolution; S13. Combine the scaling levels to convert Mercator projection coordinates into pixel coordinates, and obtain the position and number of tile data in layers.

3. The intelligent color-balancing method according to claim 1, characterized in that, The calculation of the mean and variance of the reference image tile data at each level in S2 includes the following steps: S21. Calculate the independent mean μ and variance σ² of each tile at each level in parallel. S22. Synthesize the global mean of tiles at each level by weighted averaging. and global variance .

4. The intelligent color-balancing method according to claim 3, characterized in that, The step of generating a cross-scale coefficient of variation distribution map in S3 includes: S31. Calculate the independent mean μ and variance σ² for each tile at each level; S32. Calculate the coefficient of variation (CV) for each tile based on its independent mean μ and variance σ². S23. Summarize the coefficient of variation (CV) statistics for each level of tile and generate a cross-scale coefficient of variation distribution map.

5. The intelligent color-balancing method according to claim 4, characterized in that, The coefficient of variation (CV) is expressed by the following formula: Where σ is the standard deviation and μ is the mean.

6. The intelligent color-balancing method according to claim 1, characterized in that, Multi-scale fusion is performed in time-division bands in S4.

7. The intelligent color-balancing method according to claim 1, characterized in that, In step S5, image color balancing is performed on the image to be sized using a Wallis filter. The mean and variance of the multi-scale fused reference image are used as reference values, and the color uniformity intensity is controlled by adjusting the brightness coefficient and contrast coefficient.

8. The intelligent color-balancing method according to claim 1, characterized in that, The quality assessment indicators for color uniformity evaluation include color difference analysis, radiation consistency, structural similarity, and local contrast retention.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the intelligent color balancing method as described in any one of claims 1 to 8.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent color balancing method as described in any one of claims 1 to 8.

Images