An image enhancement method based on human eye attention perception mechanism

Abstract

The application discloses an image enhancement method based on human eye attention perception mechanism and belongs to the technical field of image processing, which comprises the following steps: acquiring the gaze focus distribution data of human eyes in an original image, obtaining a visual attention distribution map based on the gaze focus distribution data, obtaining the distance weight of each pixel in the image and the gaze focus based on the visual attention distribution map, obtaining a regional human eye saliency classification matrix based on the distance weight of each pixel and the gaze focus, obtaining a hierarchical processing region map based on the regional human eye saliency classification matrix and the visual attention distribution map, and processing the original image according to the regional priority based on the hierarchical processing region map to realize the image enhancement based on the human eye attention perception mechanism. The application can effectively balance the image quality and the consumption of computing resources, realize targeted image enhancement, and improve the visual experience of users.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular relates to an image enhancement method based on the human eye's attention perception mechanism. Background Technology

[0002] Image enhancement technology, as an important branch of visual information processing, has irreplaceable value in improving image quality and optimizing user experience, especially in scenarios such as artificial intelligence, surveillance systems, and virtual reality, where its role is particularly crucial. By enhancing image details, the readability of information and the effectiveness of subsequent intelligent applications can be significantly improved; therefore, research in this field has always attracted much attention.

[0003] However, current mainstream image enhancement methods often suffer from a common deficiency: a lack of in-depth consideration of the user's subjective perception. Many techniques tend to process the entire image uniformly, ignoring the differences in human visual attention across different regions. This approach can lead to uneven resource allocation, with important areas not being adequately optimized while less important areas are over-processed, thus affecting the overall image performance and real-time capabilities.

[0004] Against this backdrop, researching how to integrate human visual attention mechanisms to guide image enhancement has become a pressing challenge. The core challenge lies in accurately capturing the focus of human attention when observing an image and transforming this focus into a criterion for judging the salience of image regions to the human eye. Due to the dynamic nature and individual differences of the human eye's gaze point, relying solely on static models makes it difficult to accurately locate the region of interest. This uncertainty in location further complicates the design of the intensity distribution for enhancement processing, leading to a discrepancy between the detail presentation in key areas and user needs.

[0005] Therefore, the key problem solved by this invention is how to construct a dynamic and intelligent enhancement processing framework based on human eye gaze behavior to achieve high-precision optimization of the area near the gaze point and gradually adjust the enhancement intensity as the distance increases. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes an image enhancement method based on the human eye's attention perception mechanism, thereby resolving the issues present in the prior art.

[0007] To achieve the above objectives, the present invention provides an image enhancement method based on the human eye's attention perception mechanism, comprising:

[0008] Obtain the gaze focus distribution data of the human eyes in the original image, and obtain the visual attention distribution map based on the gaze focus distribution data;

[0009] Based on the visual attention distribution map, the distance weight between each pixel in the image and the gaze focus is obtained, and the regional human eye saliency classification matrix is ​​obtained based on the distance weight between each pixel and the gaze focus.

[0010] Based on the region's human eye saliency classification matrix and the visual attention distribution map, a layered processing region map is obtained;

[0011] Based on the layered processing area map, the original image is processed according to the area priority to achieve image enhancement based on the human eye's attention perception mechanism.

[0012] Optionally, the process of obtaining a visual attention distribution map based on fixation focus distribution data includes:

[0013] An initial visual attention distribution map is constructed based on gaze focus distribution data; it is determined whether the gaze focus distribution data is missing or abnormal. If so, data interpolation is used to fill in the missing data to obtain a complete focus distribution dataset; features are extracted from the complete focus distribution dataset to obtain visual attention distribution features under different image scenes; if there are regions in the initial visual attention distribution map where the clustering degree of the visual attention distribution features reaches a preset condition, the regions where the clustering degree reaches the preset condition are weighted to enhance the initial visual attention distribution map and obtain a visual attention distribution map.

[0014] Optionally, the process of obtaining the distance weight between each pixel in the image and the gaze focus based on the visual attention distribution map, and obtaining the region eye saliency classification matrix based on the distance weight between each pixel and the gaze focus, includes:

[0015] Based on the visual attention distribution map, the position data of each pixel in the image is obtained. Based on the position data, the spatial distance between each pixel and the focal data is determined, and a distance weight distribution is obtained. Based on a preset weight threshold and the distance weight distribution, the region saliency classification result is obtained. Based on the region saliency classification result, regions that meet the human eye saliency requirements are extracted and their features are enhanced to obtain an enhanced region feature distribution. Through the enhanced region feature distribution, the matching degree between the spatial distance of each region and the focal data is calculated. The priority level of the matching degree clustered regions is adjusted, and the priority distribution is updated. Based on the updated priority distribution, structured data of the hierarchical matrix is ​​generated, and the human eye saliency ranking of each region in the overall distribution is determined to obtain the region human eye saliency hierarchical matrix.

[0016] Optionally, the process of obtaining a hierarchical processing region map based on the region's human eye saliency grading matrix and the visual attention distribution map includes:

[0017] The region saliency classification matrix is ​​fused with the visual attention distribution map to obtain a region saliency distribution map; the region saliency distribution map is classified according to a preset saliency threshold to obtain region classification results; the spatial distribution of each classification region is mapped based on the region classification results to obtain the region structure distribution and transform it into a visual distribution map to obtain an initial region map; a correlation rule between resource allocation and region saliency is constructed, resource allocation is performed based on the correlation rule, and the allocation scheme is integrated into the initial region map to obtain a hierarchical region map.

[0018] Optionally, the process of processing the original image according to regional priority based on the layered processing region map includes:

[0019] Based on the layered processing area map, regions with priority exceeding a threshold are obtained and detail enhancement processing is performed to obtain first image data; based on the first image data, regions with priority not exceeding the threshold are subjected to progressive smoothing processing to obtain second image data; the difference values ​​between regions in the second image data are obtained, and based on the difference values, it is determined whether boundary transition enhancement processing is performed. If boundary transition enhancement processing is performed, third image data is obtained; based on the third image data, an optimization demand distribution map is obtained, and local adaptive enhancement is performed based on the optimization demand distribution map to achieve image enhancement based on the human eye attention perception mechanism.

[0020] Optionally, the process of obtaining the first image data includes:

[0021] Based on the layered processing area map, regions with priority exceeding a threshold are segmented to obtain high-priority regions; the original image data corresponding to the high-priority regions is obtained from the original image, and the detailed feature distribution of the original image data is obtained; based on the detailed feature distribution, the original image data is enhanced using a histogram equalization method until the pixel distribution uniformity meets a preset condition; the enhanced original image data is compared with other regions in the original image, and if the difference in the comparison data is higher than a preset threshold, further optimization is performed to finally obtain the first image data.

[0022] Optionally, the process of obtaining the second image data includes:

[0023] Based on the layered processing region map and the first image data, region boundaries are segmented for regions whose priority does not exceed the threshold to obtain low-priority regions; based on the human visual salience of each region in the layered processing region map, different smoothing intensities are set for the low-priority regions, progressive smoothing processing is performed, and brightness compensation is applied until the brightness distribution uniformity meets the requirements; wherein, the resource usage information of the current task is obtained, and the layered processing region map is updated according to the resource usage information of the current task.

[0024] Optionally, the process of obtaining third image data includes:

[0025] Obtain the differential distribution feature information of the second image data, locate the boundary region based on the differential distribution feature information and the edge detection algorithm, and obtain the boundary region distribution information; perform boundary transition enhancement based on the boundary region distribution information to obtain the third image data.

[0026] Optionally, the texture distribution and color distribution of the third image data are scanned and detected to obtain areas with insufficient detail and mark them to obtain an optimization requirement distribution map.

[0027] Optionally, the process of performing local adaptive enhancement based on the optimized demand distribution map includes:

[0028] Based on the layered processing area map, the priority of each region in the optimization demand distribution map is obtained. The optimization demand distribution map is then layered according to the priority of each region to obtain the layered region data. Pixel enhancement is then performed on the layered region data to achieve image enhancement based on the human eye attention perception mechanism.

[0029] Compared with the prior art, the present invention has the following advantages and technical effects:

[0030] This invention acquires the user's gaze focus distribution from a pre-established eye-tracking database, forming a visual attention distribution map, and calculates a regional human eye saliency ranking matrix. Based on the ranking matrix, this invention performs detail enhancement on high-priority regions, smoothing on low-priority regions, and transition enhancement on boundary regions to maintain overall image consistency. Combining a user perception feedback database, this invention also performs local adaptive enhancement on regions with insufficient detail, and finally fine-tunes the global image to obtain the final enhanced image. This invention effectively balances image quality and computational resource consumption, achieving targeted image enhancement and improving the user's visual experience. Attached Figure Description

[0031] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0032] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0033] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0034] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0035] Example 1

[0036] like Figure 1 As shown, this embodiment provides an image enhancement method based on the human eye's attention perception mechanism, including:

[0037] Obtain the gaze focus distribution data of the human eyes in the original image, and obtain the visual attention distribution map based on the gaze focus distribution data;

[0038] As a specific implementation method, the process of obtaining a visual attention distribution map based on fixation focus distribution data includes:

[0039] An initial visual attention distribution map is constructed based on the gaze focus distribution data. If any gaps or anomalies exist in the gaze focus distribution data, data interpolation is used to fill them in, resulting in a complete focus distribution dataset. Feature extraction is performed on the complete focus distribution dataset to obtain visual attention distribution features for different image scenarios. If there are regions in the initial visual attention distribution map where the clustering degree of visual attention distribution features reaches a preset condition, these regions are weighted to enhance the initial visual attention distribution map, thus obtaining the final visual attention distribution map.

[0040] For example, when studying the distribution of users' visual attention to an image scene, a pre-built eye-tracking database can be used to obtain users' gaze focus data in different scenes. Suppose we are analyzing a set of indoor scene images, and the database records the gaze point distribution of 100 users in each image, but some areas have missing data because users are not looking at them. In this case, data interpolation methods, such as filling in the missing data based on the average of neighboring points, can be used to ensure the completeness of the focus distribution dataset. This method can effectively reduce the impact of incomplete data on subsequent analysis.

[0041] For example, in the feature extraction stage, a convolutional neural network model is applied to process the complete focus distribution dataset. Assuming the input is an indoor scene image, the model extracts features from regions where the user's gaze is concentrated through multiple convolutional operations. For instance, the gaze density of a certain decorative item in the image might be 80%, significantly higher than the 20% of other regions. This significant clustering indicates that the region is more attractive to the user. For such clustered regions, weighting can be applied, such as increasing the attention weight of that region by 1.5 times, generating an enhanced attention distribution map to highlight the salience of key areas in the human eye.

[0042] Based on the visual attention distribution map, the distance weight between each pixel in the image and the gaze focus is obtained, and the regional human eye saliency classification matrix is ​​obtained based on the distance weight between each pixel and the gaze focus.

[0043] As a specific implementation method, the process of obtaining the distance weight between each pixel in the image and the gaze focus based on the visual attention distribution map, and obtaining the regional human eye saliency classification matrix based on the distance weight between each pixel and the gaze focus includes:

[0044] The location data of each pixel in the image is obtained based on the visual attention distribution map. The spatial distance between each pixel and the focal data is determined based on the location data, and the distance weight distribution is obtained. The region saliency classification result is obtained based on the preset weight threshold and the distance weight distribution. Based on the region saliency classification result, regions that meet the human eye saliency requirements are extracted and their features are enhanced to obtain the enhanced region feature distribution. Through the enhanced region feature distribution, the matching degree between the spatial distance of each region and the focal data is calculated. The priority level of the matching degree clustered regions is adjusted and the priority distribution is updated. Based on the updated priority distribution, the structured data of the hierarchical matrix is ​​generated and the human eye saliency ranking of each region in the overall distribution is determined to obtain the region human eye saliency hierarchical matrix.

[0045] For example, when processing data related to visual attention distribution maps, one can first extract information about the location of each pixel in the image and analyze its spatial relationship with the focus data. Assuming an image resolution of 1920x1080, with focus data concentrated in the central region, a distance-weighted distribution map can be initially constructed by calculating the distance of each pixel to the central focus. Pixels that are closer have higher weight values; for example, the weight in the central region might be 0.9, while the weight in the edge regions might only be 0.2. This method helps to quickly distinguish the correlation between different regions in an image.

[0046] In one possible implementation, a threshold, such as 0.5, can be set for the initial distance weight distribution. Pixels with weights below this value are classified as low visual saliency regions. For example, pixels in edge regions often have weights below 0.5 and are therefore classified as low visual saliency regions, while the central region is classified as a high visual saliency region. This classification method provides a basic regional distribution map for subsequent analysis.

[0047] Feature enhancement for highly salient regions can be achieved through in-depth analysis of the texture and color information of these regions using convolutional neural network models. Assuming the central region contains a salient object, its feature distribution becomes more prominent after model processing, enhancing its recognizability within the overall image. This enhancement helps to more accurately identify key regions.

[0048] In one possible implementation, when calculating the matching degree between the regional feature distribution and the focal data, the clustering of pixels within areas of high human visual saliency can be analyzed. If the matching degree of the central region reaches over 80%, while that of the surrounding regions is only 30%, the central region is prioritized and its human visual saliency ranking is improved. This adjustment ensures the relevance of subsequent analyses.

[0049] When generating the grading matrix, the image can be divided into multiple sub-regions, each assigned a different grade value based on priority distribution. For example, the central region might be assigned grade 5, and the edge regions grade 1. This structured data facilitates subsequent processing and analysis. Next, data mapping techniques are used to fuse the matrix information with the original distribution map, generating the final grading result.

[0050] Based on the regional human eye saliency classification matrix and visual attention distribution map, a layered processing region map is obtained;

[0051] As a specific implementation method, the process of obtaining a hierarchical processing region map based on the regional human eye saliency grading matrix and visual attention distribution map includes:

[0052] The regional human eye saliency classification matrix is ​​fused with the visual attention distribution map to obtain a regional human eye saliency distribution map. The regional human eye saliency map is classified according to a preset human eye saliency threshold to obtain regional classification results. Based on the regional classification results, the spatial distribution of each classification region is mapped to obtain the regional structure distribution and transformed into a visual distribution map to obtain an initial regional map. The association rules between resource allocation and regional human eye saliency are constructed, and resource allocation is performed based on the association rules. The allocation scheme is integrated into the initial regional map to obtain a hierarchical regional map.

[0053] Specifically, if the distribution data of a certain region exceeds a threshold, it is classified as high priority, and the region classification result after hierarchical classification is determined. Based on the hierarchical region classification result, a hierarchical processing framework is constructed for the hierarchical processing business logic, mapping the spatial distribution of high-priority and low-priority corresponding regions to obtain the hierarchical processing region structure distribution.

[0054] Specifically, in the requirement to distinguish between high and low priority, when applying a preset threshold for boundary division, a weight threshold of 0.7 can be set, with areas exceeding this value classified as high-priority areas. For example, if the weight value of the image center region is 0.85, it is classified as a high-priority region, while the weight value of the edge regions is 0.3, and they are classified as low-priority regions. This division method facilitates the rationality of subsequent resource allocation.

[0055] Specifically, the spatial distribution of high-priority and low-priority regions is mapped into a hierarchical structure. Assuming the high-priority region at the image center occupies 30% of the total area, it can be stored hierarchically with other regions using spatial coordinates, forming a clear regional structure. This framework facilitates rapid location of key regions during subsequent processing.

[0056] In regional map construction, spatial mapping techniques can transform structural distributions into visual distribution maps. For example, using color coding, high-priority areas are marked in red, and low-priority areas in blue, forming an initial regional map. This visualization method facilitates an intuitive understanding of regional distribution.

[0057] Regarding the correlation between resource allocation and priority, if the central area is designated as a high-priority area, then processing resources will be allocated to it first. Assuming a total resource of 100 units, 70 units will be allocated to the high-priority area, and 30 units to the low-priority area. This priority sequence ensures that critical areas receive sufficient processing.

[0058] Based on the hierarchical processing of regional maps, the original image is processed according to regional priority to achieve image enhancement based on the human eye's attention perception mechanism.

[0059] As a specific implementation method, the process of processing the original image according to regional priority based on the layered processing region map includes:

[0060] The system firstly obtains image data by acquiring regions with priority exceeding a threshold through hierarchical processing of the region map and performing detail enhancement on these regions. Secondly, it obtains image data by progressively smoothing regions with priority below the threshold based on the first image data. Thirdly, it acquires the difference values ​​between regions in the second image data and determines whether to perform boundary transition enhancement based on these values. If boundary transition enhancement is performed, third image data is obtained. Finally, it obtains an optimization requirement distribution map based on the third image data and performs local adaptive enhancement based on this map, achieving image enhancement based on the human eye's attention perception mechanism.

[0061] As one specific implementation method, the process of obtaining the first image data includes:

[0062] Based on the hierarchical processing of the region map, regions with priority exceeding a threshold are segmented to obtain high-priority regions. The original image data corresponding to the high-priority regions is obtained from the original image, and the detailed feature distribution of the original image data is obtained. Based on the detailed feature distribution, the histogram equalization method is used to enhance the original image data until the pixel distribution uniformity meets the preset conditions. The enhanced original image data is compared with other regions in the original image. If the difference between the comparison data is higher than the preset threshold, further optimization is carried out to finally obtain the first image data.

[0063] Specifically, by processing the regional map in layers, spatial analysis tools are used to finely segment the boundaries of high-priority areas based on their distribution characteristics, obtaining the precise range distribution of these areas. Based on this precise range distribution, and addressing the business requirement of detail enhancement, the original image data within these areas is acquired. Pre-defined filtering tools are applied to initially extract image details, determining the distribution of detail features. Using this detail feature distribution, and for the specific implementation of the enhancement operation, histogram equalization in image enhancement algorithms is used to adjust the brightness and contrast of detail features, obtaining the enhanced initial image data. Based on the enhanced initial image data, and for the business logic of intensity adjustment, the pixel distribution characteristics of the image data are acquired. If the uniformity of the pixel distribution is lower than a preset threshold, intensity compensation is applied to local areas to determine the image balance state after intensity adjustment. Based on the image balance state after intensity adjustment, and for the integration requirement of region segmentation and processing priorities, comparison data between high-priority areas and other areas is acquired. If the difference in the comparison data is higher than a preset threshold, secondary detail optimization is performed on the high-priority areas, ultimately obtaining the first image data.

[0064] For example, when acquiring raw image data for precise range distribution of high-priority regions and performing detail extraction, filtering tools can be used to preliminarily process the texture and edge information in the image. Assuming the image data contains roads and vehicles, filtering tools can highlight vehicle outlines and road markings, forming a detailed feature distribution map, providing basic data support for subsequent enhancement.

[0065] For example, when enhancing images based on the distribution of detailed features, histogram equalization can adjust brightness and contrast. Suppose some areas in the original image are blurry due to insufficient light; this method can increase the brightness value from an average of 50 to 80, ensuring that details are more clearly visible. This adjustment significantly improves the accuracy of subsequent region recognition.

[0066] For the intensity adjustment logic, if the pixel distribution uniformity is lower than the preset threshold of 0.6, intensity compensation is performed on local areas. For example, if pixel values ​​in a certain area are concentrated in the low-brightness range, compensation can improve the overall uniformity to above 0.75, thereby improving image balance and providing more reliable data for regional comparison.

[0067] As one specific implementation method, the process of obtaining the second image data includes:

[0068] Based on the hierarchical processing region map and the first image data, the regions whose priority does not exceed the threshold are segmented to obtain low-priority regions. Based on the human visual salience of each region in the hierarchical processing region map, different smoothing intensities are set for the low-priority regions, and progressive smoothing processing and brightness compensation are performed until the brightness distribution uniformity meets the requirements. In this process, the resource usage information of the current task is obtained, and the hierarchical processing region map is updated according to the resource usage information of the current task.

[0069] Specifically, using the first image data, regions with priorities below a threshold are segmented to obtain low-priority regions. For the original image data of these low-priority regions, progressive smoothing is applied to initially adjust the pixel distribution within the regions, yielding smoothed intermediate image data. Based on the smoothed intermediate image data and considering the overall visual balance requirements, the brightness distribution characteristics of the image data are obtained. If the uniformity of the brightness distribution is lower than a preset threshold, brightness compensation is performed on the local areas to determine the compensated image data. Based on the compensated image data, resource allocation is optimized by obtaining the resource usage of the current processing task. If resource usage exceeds a preset limit, processing operations are prioritized to obtain an adjusted resource allocation scheme. For the processing operations in the low-priority regions, a dynamic adjustment tool is used to control the intensity of the smoothing operation in layers, resulting in optimized second image data.

[0070] Applying progressive smoothing techniques to image data in low-priority areas can be understood as adjusting the distribution by gradually reducing the degree of abrupt changes between pixels. Assuming the original image data contains noisy pixels, the smoothing technique performs mean averaging in 3x3 pixel units, gradually reducing abrupt change areas to obtain intermediate image data. This process effectively improves visual smoothness.

[0071] In brightness distribution characteristic analysis, if the uniformity is below a preset threshold, such as 0.6, local brightness compensation is required. For example, if the brightness value of a certain area is concentrated between 20 and 50, compensation can boost the darker areas to 60 to ensure a more balanced overall brightness distribution. This adjustment helps improve the visualization quality of the image.

[0072] To optimize computing resource allocation, adjustments can be made by monitoring the resource usage of current tasks. If a task consumes more than 80% of memory, priority sorting will postpone processing tasks in lower-priority areas, freeing up resources for higher-priority tasks. This dynamic allocation ensures system stability.

[0073] In the application of dynamic adjustment tools, the intensity of smoothing operations can be controlled hierarchically, with different smoothing parameters set according to the salience of different regions to the human eye. For example, setting the smoothing intensity to 0.3 for edge regions and 0.5 for the central transition region will result in a more natural transition effect in the second image data. This hierarchical control optimizes resource utilization efficiency.

[0074] As one specific implementation method, the process of obtaining third image data includes:

[0075] The differential distribution feature information of the second image data is obtained. Based on the differential distribution feature information and the edge detection algorithm, the boundary region is located to obtain the boundary region distribution information. Based on the boundary region distribution information, the boundary transition is enhanced to obtain the third image data.

[0076] Specifically, using the second image data, a pre-defined analysis tool is employed to scan the distribution of regional differences, acquiring feature information and determining the initial range of differences. Based on this initial range, an edge detection algorithm is applied to locate the boundary regions, obtaining precise distribution information and resulting in the located boundary data. Using this located boundary data, a progressive enhancement algorithm is used to adjust the pixel distribution of the boundary regions to meet the need for over-enhancement, acquiring enhanced intermediate data and determining the adjusted image features, thus obtaining the third image data.

[0077] As a specific implementation method, the texture distribution and color distribution of the third image data are scanned and detected to obtain areas with insufficient detail and mark them to obtain an optimization requirement distribution map.

[0078] As a specific implementation method, the process of local adaptive enhancement based on the optimized demand distribution map includes:

[0079] Based on the hierarchical processing of the regional map, the priority of each region in the optimization demand distribution map is obtained. The optimization demand distribution map is then layered according to the priority of each region to obtain the layered regional data. Pixel enhancement is then performed on the layered regional data to achieve image enhancement based on the human eye attention perception mechanism.

[0080] Specifically, for areas with insufficient detail in the optimization requirement distribution map, a pre-set image analysis tool is used to scan the marked areas to obtain feature distribution information of the deficient areas. The priority range for local enhancement is determined based on the layered processing area map. According to the priority range of local enhancement, for adaptive adjustment needs, an image adjustment tool is applied to perform layered processing on the marked areas, obtaining layered area data to determine the initial direction of enhancement processing. If the feature distribution of the layered area data does not match a preset threshold, the weight allocation of specific marked areas is adjusted, obtaining the adjusted weight distribution information to determine the specific adaptive enhancement scheme. Based on the adjusted weight distribution information, for image content optimization, a convolutional neural network in a deep learning model is used to extract detailed features from the marked areas, obtaining extracted detailed feature data to determine the applicable scope of enhancement processing. If the detailed feature data deviates from the preset sharpness standard, local pixel-level enhancement is performed on the marked areas, obtaining enhanced area image information to determine the final adjusted detail content. The adjusted detail content is fused with the overall image content to obtain fused image data, which is the final image output, achieving image enhancement based on the human eye's attention perception mechanism.

[0081] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An image enhancement method based on the human eye's attention perception mechanism, characterized in that, Includes the following steps: Obtain the gaze focus distribution data of the human eyes in the original image, and obtain the visual attention distribution map based on the gaze focus distribution data; Based on the visual attention distribution map, the distance weight between each pixel in the image and the gaze focus is obtained, and the regional human eye saliency classification matrix is ​​obtained based on the distance weight between each pixel and the gaze focus. Based on the region's human eye saliency classification matrix and the visual attention distribution map, a layered processing region map is obtained; Based on the layered processing area map, the original image is processed according to the area priority to achieve image enhancement based on the human eye's attention perception mechanism; The process of obtaining the distance weight between each pixel in the image and the gaze focus based on the visual attention distribution map, and obtaining the region saliency classification matrix based on the distance weight between each pixel and the gaze focus, includes: Based on the visual attention distribution map, the position data of each pixel in the image is obtained. Based on the position data, the spatial distance between each pixel and the focal data is determined, and a distance weight distribution is obtained. Based on a preset weight threshold and the distance weight distribution, the region saliency classification result is obtained. Based on the region saliency classification result, regions that meet the human eye saliency requirements are extracted and their features are enhanced to obtain an enhanced region feature distribution. Through the enhanced region feature distribution, the matching degree between the spatial distance of each region and the focal data is calculated. The priority level of the matching degree clustered regions is adjusted, and the priority distribution is updated. Based on the updated priority distribution, structured data of the hierarchical matrix is ​​generated, and the human eye saliency ranking of each region in the overall distribution is determined to obtain the region human eye saliency hierarchical matrix.

2. The image enhancement method based on the human eye's attention perception mechanism according to claim 1, characterized in that, The process of obtaining a visual attention distribution map based on fixation focus distribution data includes: An initial visual attention distribution map is constructed based on gaze focus distribution data; it is determined whether the gaze focus distribution data is missing or abnormal. If so, data interpolation is used to fill in the missing data to obtain a complete focus distribution dataset; features are extracted from the complete focus distribution dataset to obtain visual attention distribution features under different image scenes; if there are regions in the initial visual attention distribution map where the clustering degree of the visual attention distribution features reaches a preset condition, the regions where the clustering degree reaches the preset condition are weighted to enhance the initial visual attention distribution map and obtain a visual attention distribution map.

3. The image enhancement method based on the human eye's attention perception mechanism according to claim 1, characterized in that, The process of obtaining a layered processing region map based on the region's human eye saliency grading matrix and the visual attention distribution map includes: The region saliency classification matrix is ​​fused with the visual attention distribution map to obtain a region saliency distribution map; the region saliency distribution map is classified according to a preset saliency threshold to obtain region classification results; the spatial distribution of each classification region is mapped based on the region classification results to obtain the region structure distribution and transform it into a visual distribution map to obtain an initial region map; a correlation rule between resource allocation and region saliency is constructed, resource allocation is performed based on the correlation rule, and the allocation scheme is integrated into the initial region map to obtain a hierarchical region map.

4. The image enhancement method based on the human eye's attention perception mechanism according to claim 3, characterized in that, The process of processing the original image according to regional priority based on the layered processing region map includes: Based on the layered processing area map, regions with priority exceeding a threshold are obtained and detail enhancement processing is performed to obtain first image data; based on the first image data, regions with priority not exceeding the threshold are subjected to progressive smoothing processing to obtain second image data; the difference values ​​between regions in the second image data are obtained, and based on the difference values, it is determined whether boundary transition enhancement processing is performed. If boundary transition enhancement processing is performed, third image data is obtained; based on the third image data, an optimization demand distribution map is obtained, and local adaptive enhancement is performed based on the optimization demand distribution map to achieve image enhancement based on the human eye attention perception mechanism.

5. The image enhancement method based on the human eye's attention perception mechanism according to claim 4, characterized in that, The process of obtaining the first image data includes: Based on the layered processing area map, regions with priority exceeding a threshold are segmented to obtain high-priority regions; the original image data corresponding to the high-priority regions is obtained from the original image, and the detailed feature distribution of the original image data is obtained; based on the detailed feature distribution, the original image data is enhanced using a histogram equalization method until the pixel distribution uniformity meets a preset condition; the enhanced original image data is compared with other regions in the original image, and if the difference in the comparison data is higher than a preset threshold, further optimization is performed to finally obtain the first image data.

6. The image enhancement method based on the human eye's attention perception mechanism according to claim 4, characterized in that, The process of obtaining the second image data includes: Based on the layered processing region map and the first image data, region boundaries are segmented for regions whose priority does not exceed the threshold to obtain low-priority regions; based on the human visual salience of each region in the layered processing region map, different smoothing intensities are set for the low-priority regions, progressive smoothing processing is performed, and brightness compensation is applied until the brightness distribution uniformity meets the requirements; wherein, the resource usage information of the current task is obtained, and the layered processing region map is updated according to the resource usage information of the current task.

7. The image enhancement method based on the human eye's attention perception mechanism according to claim 4, characterized in that, The process of obtaining third image data includes: Obtain the differential distribution feature information of the second image data, locate the boundary region based on the differential distribution feature information and the edge detection algorithm, and obtain the boundary region distribution information; perform boundary transition enhancement based on the boundary region distribution information to obtain the third image data.

8. The image enhancement method based on the human eye's attention perception mechanism according to claim 4, characterized in that, The texture and color distribution of the third image data are scanned and detected to obtain areas with insufficient detail and mark them to obtain an optimization requirement distribution map.

9. The image enhancement method based on the human eye's attention perception mechanism according to claim 4, characterized in that, The process of local adaptive enhancement based on the aforementioned optimized demand distribution map includes: Based on the layered processing area map, the priority of each region in the optimization demand distribution map is obtained. The optimization demand distribution map is then layered according to the priority of each region to obtain the layered region data. Pixel enhancement is then performed on the layered region data to achieve image enhancement based on the human eye attention perception mechanism.

Images