A large-scale remote sensing image color correction method based on block local histogram matching

Abstract

The application discloses a large-scale remote sensing image color correction method based on block local histogram matching, belongs to the technical field of remote sensing image processing, and aims at the problems of low calculation efficiency, neglect of local features and difficulty in maintaining overall color tone when existing methods process large-scale orthographic images. The correction is realized through five steps of adaptive image blocking, multi-scale statistical feature extraction, hierarchical nonlinear mapping model, high-precision seamless splicing and parallel computing framework optimization. The method can efficiently process massive data, takes into account local correction accuracy and overall color tone consistency, improves robustness through intelligent parameter adaptive adjustment, significantly eliminates image color differences, is suitable for multiple fields such as land resource investigation and city planning, and provides a high-quality solution for large-scale image splicing.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and specifically to a method for color correction of large-scale remote sensing images based on block-based local histogram matching. Background Technology

[0002] With the rapid development of remote sensing technology, large-scale orthorectified remote sensing imagery has been widely used in many fields such as land resource surveys, environmental monitoring, urban planning, agricultural management, and disaster assessment. These applications provide us with valuable Earth observation data, supporting various spatial decision-making and scientific research. However, during the mosaicking and processing of large-scale remote sensing orthorectified images, due to the influence of various complex factors, there are often significant color differences between images acquired at different times or by different sensors. This problem is particularly prominent in the orthorectified image mosaicking process, which seriously affects the visual quality of the images and the accuracy of subsequent analysis.

[0003] The color discrepancies in remote sensing imagery stem primarily from the interplay of multiple factors. First, changes in atmospheric conditions significantly impact illumination. Factors such as cloud cover, aerosol content, and water vapor content alter the intensity and spectral distribution of light received by the Earth's surface. These variations, especially at the boundaries of images taken at different times or locations, can easily create noticeable color jumps during image stitching, affecting the overall visual effect. Second, the differences in spectral response characteristics among different remote sensing sensors are also a major cause of color inconsistencies. Because various remote sensing sensors possess different spectral response curves, the spectral characteristics of the same ground feature can vary drastically across different sensors. This leads to inconsistent color tones between different areas, particularly during large-scale image stitching involving data from multiple sensors. Furthermore, factors such as imaging time, observation angle, and seasonal variations also affect the reflectivity and shadow effects of ground features. Especially in terrain-complex areas like mountains and cities, changes in solar altitude angle and observation angle can cause fluctuations in radiance values, further exacerbating color differences in the images.

[0004] With the rapid increase in remote sensing image data, traditional color correction methods face technical challenges such as low computational efficiency and memory overflow when processing large-scale data. Especially under the requirements of rapid mosaicking and real-time updates, the processing capacity of existing methods is stretched thin. Furthermore, during orthorectification, the geometric distortion correction of images due to terrain undulations can introduce additional color changes, particularly in complex terrain areas such as mountains, where the impact is more significant. The diversity of land cover types (such as vegetation, water bodies, and buildings) and their variations in spectral characteristics further complicate image mosaicking. In this environment, the spectral differences between different types of land cover pose a significant challenge to color correction, especially in areas where land cover types frequently change.

[0005] Furthermore, long-term use of sensors may lead to performance degradation and calibration errors. These problems accumulate over time, potentially causing systematic color deviations in images captured at different times. Traditional atmospheric correction algorithms often perform inconsistently when dealing with these issues, failing to meet the processing requirements of large-scale image data, ultimately resulting in color differences remaining in the corrected images. Therefore, effectively addressing these technical challenges affecting color consistency during large-scale remote sensing image stitching becomes crucial for improving image quality and subsequent analysis accuracy.

[0006] Color differences in orthophoto mosaicking not only affect visual aesthetics, but more importantly, they can lead to errors in subsequent quantitative analysis and information extraction, impacting the scientific rigor and accuracy of decision-making. For example, in applications such as change detection and land use classification, color inconsistencies can result in erroneous judgments and classifications.

[0007] Currently used color correction methods include global histogram matching, linear stretching, radiometric calibration, and feature-based color transformation. While global histogram matching can adjust the histogram distribution of the image to be corrected as a whole, it is computationally intensive when processing large-scale orthophotos and ignores local features, easily leading to overcorrection and loss of detail. Linear stretching is simple and fast, but it struggles to handle non-linear color differences and cannot maintain the overall tone style after mosaicking. Radiometric calibration, based on the radiative transfer model, can theoretically yield relatively accurate results, but it requires a large amount of ground-based measurement data, making it difficult to implement in large-scale orthophoto mosaicking and computationally complex. Feature-based color transformation extracts image feature points for color mapping, but its effectiveness is poor in areas with indistinct ground features. These methods generally suffer from shortcomings when processing large-scale orthophoto mosaics, such as low computational efficiency, neglect of local features, inability to maintain the overall tone style, high computational resource requirements, and lack of targeted processing for the unique deformations and color changes of orthophotos. They are difficult to meet the needs of large-scale data processing, real-time updates, and rapid processing, and also affect the visual consistency and correction accuracy of the mosaicked large-scale images, especially in areas with complex terrain or diverse land cover types.

[0008] Therefore, developing an efficient, accurate, and comprehensive color correction method for large-scale orthophotos that maintains the overall tonal style is of great significance for improving the quality and application value of remote sensing images. Summary of the Invention

[0009] This invention proposes a color correction method for large-scale remote sensing images based on block-based local histogram matching, the method comprising the following steps: S1. Adaptive image segmentation and intelligent overlapping region determination; S2. Multi-scale block statistical feature extraction and analysis; S3. Construction of advanced hierarchical nonlinear mapping models, with an innovative two-level mapping model architecture, including macroscopic block level and microscopic local window level; S4. High-precision seamless stitching and local histogram matching technology, improved cumulative distribution function (CDF) calculation method, and designed smooth transition algorithm for overlapping areas; S5. Optimize the high-performance parallel computing framework, and design and implement a hybrid parallel processing framework based on multi-core CPUs and GPUs.

[0010] The adaptive image segmentation and intelligent overlap region determination in step S1 includes the following steps: S1.1 dynamically determines the optimal block size based on image features and available computing resources. This adaptive mechanism can find the best balance between efficiency and accuracy in image processing tasks of different scales and complexities. S1.2 uses an intelligent algorithm to set adaptive overlapping areas between adjacent blocks, which not only ensures a high degree of continuity in color transitions, but also lays the foundation for seamless splicing in the future. S1.3 uses an overlapping region analysis algorithm to accurately calculate and optimize the overlapping range of adjacent blocks, thereby minimizing computational overhead while ensuring correction quality. The formula for calculating the optimal block size in step S1.1 is as follows: ; in, To achieve the optimal block size, This represents the computational efficiency function. This represents the correction accuracy function. It is a balancing factor.

[0011] The multi-scale block statistical feature extraction and analysis in step S2 introduces an adaptive local window mechanism. The window size can be dynamically adjusted according to the image content. The basic size is set to 256×256 pixels, but it can be flexibly changed according to the actual situation.

[0012] To improve the accuracy and robustness of feature representation, this method employs an improved histogram calculation approach, which better adapts to different types of land cover features. Simultaneously, an innovative feature consistency assessment model was developed, providing a reliable reference for subsequent color correction by deeply analyzing the feature correlations of overlapping areas in adjacent blocks.

[0013] Its local features and local statistical features are calculated using the following formulas: ; in, The mean, Standard deviation For pixel values,N This represents the total number of pixels.

[0014] In the macro-block-level model of step S3, a global tone style preservation mechanism is introduced to ensure the consistency of the overall visual effect and avoid the local overcorrection problem common in traditional methods.

[0015] In the micro-local window-level model of step S3, an advanced nonlinear correction algorithm is used to achieve fine color adjustment, which can effectively handle complex lighting changes and ground reflection characteristics.

[0016] The nonlinear mapping function is: ; in, x For input pixel values, y For the corrected pixel values, a, b, c, d These are the parameters to be optimized.

[0017] In the construction of the advanced hierarchical nonlinear mapping model in step S3, this system also integrates machine learning technology to build an intelligent parameter adaptive adjustment system, which can automatically optimize the processing parameters according to different image features, greatly improving the adaptability and robustness of the method.

[0018] The improved cumulative distribution function (CDF) calculation method in step S4 significantly improves the accuracy of histogram matching.

[0019] The formula for calculating the cumulative distribution function is: ; in, Grayscale The grayscale value is The number of pixels, N This represents the total number of pixels.

[0020] The overlapping area smooth transition algorithm in step S4 achieves a truly seamless splicing effect.

[0021] The high-precision seamless splicing and local histogram matching technology in step S4 introduces a boundary continuity constraint mechanism to further ensure the naturalness of the transition between blocks, effectively eliminating splicing traces.

[0022] The high-precision seamless stitching and local histogram matching technology in step S4 also includes a global tone consistency evaluation and adjustment module. Through intelligent analysis and dynamic adjustment, the overall style is kept harmonious and unified, resulting in a high degree of visual coherence in the final stitched large-scale image.

[0023] The high-performance parallel computing framework described in step S5 has developed an efficient task scheduling algorithm to achieve dynamic load balancing of block data and ensure optimal utilization of computing resources.

[0024] The high-performance parallel computing framework described in step S5 introduces advanced memory management strategies and data access optimization techniques in terms of data management, effectively minimizing I / O bottlenecks and further improving system performance.

[0025] The high-performance parallel computing framework described in step S5, this system also integrates a distributed computing framework, providing scalable technical support for processing larger-scale data in the future.

[0026] The formula for calculating the parallel speedup ratio is as follows: ; in, For acceleration ratio, 1 represents the serial execution time. For use Parallel execution time of each processor.

[0027] Compared with the prior art, the beneficial effects of the present invention include: (1) It can efficiently process large-scale data, innovate block processing strategies, greatly improve the color and light uniformity efficiency of massive remote sensing images, and expand the application scope of the system.

[0028] (2) The tone correction effect has been optimized by adopting a hierarchical nonlinear mapping model, which takes into account both the accuracy of local correction and the consistency of overall tone style, thus solving the problem that traditional methods are difficult to balance.

[0029] (3) Achieve high-precision color and light uniformity. Through improved histogram matching technology and local feature analysis, effectively eliminate image color differences and improve overall visual consistency.

[0030] (4) Adaptive adjustment enhances robustness. Machine learning technology is integrated to build an intelligent parameter adaptive system that can automatically optimize parameters according to image features and adapt to different scenarios.

[0031] (5) High-performance computing support: Based on a hybrid parallel architecture of multi-core CPU and GPU, coupled with efficient task scheduling and memory management, it significantly improves processing efficiency and meets the needs of rapid processing of large-scale data. Attached Figure Description

[0032] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0033] Figure 1 This is an overall flowchart of the method of the present invention.

[0034] Figure 2 This is a diagram of the computational architecture of the method of the present invention.

[0035] Figure 3 The image shown is the original image before color correction, as in this embodiment.

[0036] Figure 4 The image to be stitched is shown in the example.

[0037] Figure 5 The image shown is a color-corrected version of the example.

[0038] Figure 6 This is an overall color uniformity effect diagram for the embodiment.

[0039] Figure 7 This is a partial splicing effect diagram of an embodiment. Detailed Implementation

[0040] The overall flowchart of the large-scale remote sensing image color correction method based on block local histogram matching described in this invention is as follows: Figure 1 As shown, the specific steps are as follows: S1. Adaptive image segmentation and intelligent overlapping region determination; This step first dynamically determines the optimal tile size based on image features and available computing resources. This adaptive mechanism can find the best balance between efficiency and accuracy in image processing tasks of varying scales and complexities. Next, an intelligent algorithm is used to set adaptive overlap regions between adjacent tiles, ensuring not only a high degree of continuity in color transitions but also laying the foundation for seamless stitching. During this process, the system utilizes a high-precision geographic coordinate system to accurately acquire the spatial location information of each tile, supporting subsequent geographic information correlation analysis. Furthermore, this invention has developed a specialized overlap region analysis algorithm that can accurately calculate and optimize the overlap range of adjacent tiles, thereby minimizing computational overhead while ensuring correction quality.

[0041] The formula for calculating the optimal block size: ; in, To achieve the optimal block size, This represents the computational efficiency function. This represents the correction accuracy function. It is a balancing factor.

[0042] S2. Multi-scale block statistical feature extraction and analysis; In this step, the system performs multi-level, multi-scale local statistical feature extraction on each block. This comprehensive feature analysis can capture subtle changes and local characteristics in the image. This invention introduces an adaptive local window mechanism; the window size can be dynamically adjusted according to the image content, with a base size of 256×256 pixels, but can be flexibly changed according to actual conditions. To improve the accuracy and robustness of feature representation, this method adopts an improved histogram calculation method, which can better adapt to different types of ground feature characteristics. Simultaneously, an innovative feature consistency evaluation model is developed, which provides a reliable reference for subsequent color correction by deeply analyzing the feature correlation of overlapping areas between adjacent blocks, such as... Figure 3 The image shown is the original image before color correction. Figure 4 Images to be stitched together.

[0043] Formula for calculating local statistical characteristics: ; in, The mean, Standard deviation For pixel values, N This represents the total number of pixels.

[0044] S3. Construction of advanced hierarchical nonlinear mapping models, with an innovative two-level mapping model architecture, including macroscopic block level and microscopic local window level; At the block level, a global tone style preservation mechanism is introduced to ensure the consistency of the overall visual effect and avoid the local overcorrection problem common in traditional methods. At the local window level, an advanced nonlinear correction algorithm is used to achieve fine-grained color adjustment, effectively handling complex lighting changes and ground feature reflection characteristics, such as... Figure 5 This is the image after color correction. Furthermore, this system integrates machine learning technology to construct an intelligent parameter adaptive adjustment system that can automatically optimize processing parameters based on different image features, greatly improving the adaptability and robustness of the method.

[0045] Nonlinear mapping function: ; in, x For input pixel values, y For the corrected pixel values, a, b, c, d These are the parameters to be optimized.

[0046] S4. High-precision seamless stitching and local histogram matching technology, improved cumulative distribution function (CDF) calculation method, and designed smooth transition algorithm for overlapping areas; To achieve high-quality image stitching, such as Figure 6To achieve a uniform color effect, this invention develops an improved Cumulative Distribution Function (CDF) calculation method. By dividing the image into multiple small blocks for separate processing, each block's CDF is calculated independently. Parallel computing technology is used to improve processing speed. Furthermore, a local weighted accumulation mechanism is introduced to optimize calculations based on the pixel distribution of each block, thereby significantly improving color matching accuracy and histogram matching precision. Simultaneously, a specialized algorithm for smoothing overlapping regions is designed. By dynamically adjusting the pixel weighting coefficients, it ensures natural color transitions, achieving a truly seamless stitching effect. To further ensure the naturalness of transitions between blocks, a boundary continuity constraint mechanism is introduced, effectively eliminating stitching artifacts, such as... Figure 7 This is a partial stitched image. In addition, this system has developed a global tone consistency assessment and adjustment module. Through intelligent analysis and dynamic adjustment, it ensures the overall style is harmonious and unified, resulting in a high degree of visual coherence in the final stitched large-scale image.

[0047] Formula for calculating cumulative distribution function: ; in, Grayscale The grayscale value is The number of pixels, N This represents the total number of pixels.

[0048] S5. Optimize the high-performance parallel computing framework, and design and implement a hybrid parallel processing framework based on multi-core CPUs and GPUs; To address the challenges of large-scale data processing, this invention designs and implements a hybrid parallel processing architecture based on multi-core CPUs and GPUs, as shown in the following diagram. Figure 2 As shown, this advanced computing framework fully leverages the parallel computing capabilities of modern hardware, significantly improving processing efficiency. Simultaneously, an efficient task scheduling algorithm has been developed to achieve dynamic load balancing of segmented data, ensuring optimal utilization of computing resources. In terms of data management, advanced memory management strategies and data access optimization techniques have been introduced, effectively minimizing I / O bottlenecks and further enhancing system performance. Furthermore, this system integrates a distributed computing framework, providing scalable technical support for processing larger-scale data in the future.

[0049] Formula for calculating parallel speedup: ; in, For acceleration ratio, 1 represents the serial execution time. For use Parallel execution time of each processor.

Claims

1. A color correction method for large-scale remote sensing images based on block-based local histogram matching, characterized in that, Includes the following steps: S1. Adaptive image segmentation and intelligent overlapping region determination; S2. Multi-scale block statistical feature extraction and analysis; S3. Construction of advanced hierarchical nonlinear mapping models, with an innovative two-level mapping model architecture, including macroscopic block level and microscopic local window level; S4. High-precision seamless stitching and local histogram matching technology, improved cumulative distribution function calculation method, and designed smooth transition algorithm for overlapping areas; S5. Optimize the high-performance parallel computing framework, and design and implement a hybrid parallel processing framework based on multi-core CPUs and GPUs.

2. The method for color correction of large-scale remote sensing images based on block-based local histogram matching according to claim 1, characterized in that, The adaptive image segmentation and intelligent overlap region determination in step S1 includes the following steps: S1.1 dynamically determines the optimal block size based on image features and available computing resources. This adaptive mechanism can find the best balance between efficiency and accuracy in image processing tasks of different scales and complexities. S1.2 uses an intelligent algorithm to set adaptive overlapping areas between adjacent blocks, which not only ensures a high degree of continuity in color transitions, but also lays the foundation for seamless splicing in the future. S1.3 uses an overlapping region analysis algorithm to accurately calculate and optimize the overlapping range of adjacent blocks, thereby minimizing computational overhead while ensuring correction quality.

3. The method for color correction of large-scale remote sensing images based on block-based local histogram matching according to claim 2, characterized in that, The formula for calculating the optimal block size in step S1.1 is as follows: ； in, To achieve the optimal block size, This represents the computational efficiency function. This represents the correction accuracy function. As a balance factor; The intelligent algorithm in step S1.2 refers to the system automatically calculating the block size and overlapping area based on the current computing resources. The system automatically determines the most suitable block size based on the complexity of the image and the available computing resources, and optimizes the overlapping area between each block. This intelligent algorithm dynamically adjusts task allocation and block strategy by monitoring the usage of system resources in real time, such as the memory usage and computing load of CPU and GPU, in order to ensure a balance between processing efficiency and accuracy.

4. The method for color correction of large-scale remote sensing images based on block-based local histogram matching according to claim 1, characterized in that, The multi-scale block statistical feature extraction and analysis in step S2 introduces an adaptive local window mechanism. The window size can be dynamically adjusted according to the image content. The basic size is set to 256×256 pixels, but it can be flexibly changed according to the actual situation. In order to improve the accuracy and robustness of feature expression, an improved histogram calculation method is adopted, which can better adapt to different types of land cover features. At the same time, an innovative feature consistency evaluation model is developed. By deeply analyzing the feature correlation of adjacent block overlapping areas, a reliable reference basis is provided for subsequent color correction. The formula for calculating its local features and local statistical features is as follows: ； in, The mean, Standard deviation, For pixel values, N This represents the total number of pixels.

5. The method for color correction of large-scale remote sensing images based on block-based local histogram matching according to claim 1, characterized in that, In the macro-block-level model of step S3, a global tone style preservation mechanism is introduced to ensure the consistency of the overall visual effect and avoid the local overcorrection problem common in traditional methods.

6. The method for color correction of large-scale remote sensing images based on block-based local histogram matching according to claim 1, characterized in that, In the micro-local window-level model of step S3, an advanced nonlinear correction algorithm is used to achieve fine color adjustment, which can effectively handle complex lighting changes and ground object reflection characteristics. The nonlinear mapping function is: ； in, x For input pixel values, y For the corrected pixel values, a, b, c, d These are the parameters to be optimized.

7. The method for color correction of large-scale remote sensing images based on block-based local histogram matching according to claim 1, characterized in that, The improved cumulative distribution function (CDF) calculation method in step S4 divides the image into multiple small blocks for processing through block-based calculation. Each block calculates the CDF independently and uses parallel computing technology to improve processing speed. In addition, a local weighted accumulation mechanism is introduced to optimize the calculation based on the pixel distribution of each block, thereby improving the accuracy of color matching and significantly improving the accuracy of histogram matching. The formula for calculating the cumulative distribution function is: ； in, Grayscale The grayscale value is The number of pixels, N This represents the total number of pixels.

8. The method for color correction of large-scale remote sensing images based on block-based local histogram matching according to claim 1, characterized in that, The overlapping area smooth transition algorithm in step S4 ensures natural color transitions by dynamically adjusting the weighting coefficients of pixels, achieving a truly seamless stitching effect. To further ensure the naturalness of the transition between blocks, a boundary continuity constraint mechanism is introduced, effectively eliminating stitching marks. A global tone consistency evaluation and adjustment module has also been developed. Through intelligent analysis and dynamic adjustment, the overall style is kept harmonious and unified, resulting in a high degree of visual coherence in the final stitched large-scale image.

9. A method for color correction of large-scale remote sensing images based on block-based local histogram matching according to claim 1, characterized in that, The high-performance parallel computing framework in step S5 has developed an efficient task scheduling algorithm to achieve dynamic load balancing of block data and ensure optimal utilization of computing resources. In terms of data management, it has introduced advanced memory management strategies and data access optimization techniques to effectively minimize I / O bottlenecks and further improve system performance. The efficient task scheduling algorithm specifically refers to a hybrid parallel computing scheduling algorithm based on CPU and GPU. This algorithm allocates computationally intensive tasks, such as CDF calculation and pixel stretching, to the GPU for parallel processing, while simpler control tasks are handled by the CPU. At the same time, the task scheduling algorithm adjusts task allocation in real time according to the current memory usage and computing load of the CPU and GPU to ensure efficient utilization of computing resources. Through a priority scheduling mechanism, important tasks are processed first, reducing computing bottlenecks and ensuring the efficiency of large-scale data processing.

10. A method for color correction of large-scale remote sensing images based on block-based local histogram matching according to claim 1, characterized in that, The high-performance parallel computing framework described in step S5, this system also integrates a distributed computing framework, providing scalable technical support for processing larger-scale data in the future. The formula for calculating the parallel speedup ratio is as follows: ； in, For acceleration ratio, 1 represents the serial execution time. For use Parallel execution time of each processor.

Images