High dynamic range image enhancement method and enhancement system based on artificial intelligence

Abstract

The present invention provides a high dynamic range image enhancement method and enhancement system based on artificial intelligence, which relates to the technical field of image processing. By extracting the global brightness features of the HDR image and the local features of the initial block, the overall brightness and local details of the image can be fully understood; then, the initial block is segmented based on the local features to obtain a plurality of uniform sub-blocks, and the sub-blocks are super-resolved through a preset super-resolution module to output high-resolution blocks, which can retain more details in the image and improve the resolution of the image at the same time; then, the output high-resolution blocks are feature fused and enhanced, which can effectively improve the contrast and clarity of the image; finally, the processed high-resolution blocks are spliced ​​to obtain a final high-resolution HDR image, which can ensure that the final image reaches a high quality in both global and local details and meet the needs of various application scenarios.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a high dynamic range image enhancement method and enhancement system based on artificial intelligence. Background Art

[0002] With the rapid development of digital imaging technology, high dynamic range (HDR) images are widely used in film, television, advertising, virtual reality and other fields due to their ability to display a wider brightness range and color depth. However, due to the large amount of HDR image data and high processing complexity, how to effectively enhance HDR images to improve the visual effect of the image has always been a research hotspot in the field of image processing.

[0003] Traditional HDR image enhancement methods are often based on global brightness adjustment or local contrast enhancement. Although these methods can improve the visual effect of the image to a certain extent, they have the following problems: First, global brightness adjustment is difficult to adapt to the large brightness difference in HDR images, which can easily lead to overexposure or underexposure; second, although local contrast enhancement methods can improve image details, they may cause edge blur or noise amplification.

[0004] Therefore, it is necessary to provide a high dynamic range image enhancement method and enhancement system based on artificial intelligence to solve the above technical problems. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a high dynamic range image enhancement method and enhancement system based on artificial intelligence, which realizes efficient enhancement of HDR images by combining the processing of global features and local features.

[0006] The present invention provides a high dynamic range image enhancement method based on artificial intelligence, and the enhancement method comprises the following steps:

[0007] S1: Acquire an HDR image, and extract a global feature of the HDR image, wherein the global feature is a brightness feature;

[0008] S2: performing region segmentation on the HDR image based on the brightness feature to obtain a plurality of uniform initial blocks corresponding to the HDR image;

[0009] S3: extracting local features of the initial block, and performing block segmentation on the initial block based on the local features to obtain a plurality of uniform sub-blocks corresponding to the initial block;

[0010] S4: inputting all sub-blocks into a preset super-resolution module for super-resolution processing, and outputting a plurality of high-resolution blocks corresponding to the sub-blocks, wherein the super-resolution module is a neural network model trained by including multiple types of sample sub-blocks;

[0011] S5: According to the global and local features of the HDR image, feature fusion and enhancement processing are performed on the output multiple high-resolution blocks, and the processed multiple high-resolution blocks are spliced ​​to obtain a final high-resolution HDR image.

[0012] Preferably, step S1 specifically includes:

[0013] S101: Preprocessing the acquired HDR image, wherein the preprocessing includes denoising and distortion correction;

[0014] S102: converting the HDR image into a brightness histogram, and assigning a brightness level to each column in the brightness histogram;

[0015] S103: traversing each brightness level in the brightness histogram, and weighted summing all brightness levels to obtain a brightness sum;

[0016] S104: Calculate the sum of all columnar heights in the brightness histogram to obtain the total number of pixels;

[0017] S105: Divide the total brightness by the total number of pixels to obtain an average brightness as a brightness feature.

[0018] Preferably, step S2 specifically includes:

[0019] S201: Determine multiple region segmentation brightness thresholds based on a statistical method;

[0020] S202: Compare the multiple region segmentation brightness thresholds with each pixel of the HDR image, and divide the HDR image into multiple uniform initial blocks according to the comparison result, wherein the initial block is represented as:

[0021] R i ={(x, y)|L(x, y)∈[T i-1 , T i )}

[0022] Among them, R i represents the i-th initial block, T i represents the brightness threshold of the i-th region segmentation, [T i-1 , T i ) represents the brightness value range of the initial block; L(x, y) represents the brightness value of the position (x, y).

[0023] Preferably, step S3 specifically includes:

[0024] S301: traverse each of the initial blocks and extract local features of the initial blocks, wherein the local features are local brightness Ll and the edge strength E, are respectively expressed as:

[0025]

[0026]

[0027] where N(x, y) represents the neighborhood around the position (x, y), I(m, n) represents the pixel value at the position (m, n), and N is the number of pixels in the neighborhood. represents the gradient strength at the position (x, y);

[0028] S302: Determine the local brightness threshold T L and the edge strength threshold T E , which are respectively expressed as:

[0029] T L = arg min T (σ 2 (L l < T) + σ 2 (L l ≥ T))

[0030] T E = arg min T (σ 2 (E < T) + σ 2 (E ≥ T))

[0031] where T represents the dynamic candidate threshold, and arg min T represents the dynamic candidate threshold T when the sum of variances is the smallest. σ 2 (L l < T) and σ 2 (L l ≥ T) respectively represent the variances of the pixel values where the local brightness L l is less than the dynamic candidate threshold T and greater than or equal to the dynamic candidate threshold T. σ 2 (E < T i ) and σ 2 (E ≥ T i ) respectively represent the variances of the pixel values where the edge strength E is less than the dynamic candidate threshold T and greater than or equal to the dynamic candidate threshold T;

[0032] S303: Segment each initial block based on the local brightness threshold T L and the edge strength threshold T E to obtain sub - blocks, where the sub - blocks are expressed as:

[0033] C j = {(x, y) ∈ B | L l ∈ [TL(j-1) , T L(j) ), E>T E}

[0034] Among them, C j Represented as the jth sub-block, [T L(j-1) , T L(j) ) represents the local brightness value range of the j-th sub-block, and B represents the initial block.

[0035] Preferably, step S4 specifically includes:

[0036] S401: passing each sub-block as input to a pre-trained neural network-based super-resolution model;

[0037] S402: performing super-resolution processing on each input sub-block using the super-resolution model, and improving the resolution of each sub-block through the learned weight and bias parameters;

[0038] S403: The super-resolution model outputs a high-resolution block corresponding to each input sub-block.

[0039] Preferably, step S5 specifically includes:

[0040] S501: combining the global feature and the local feature of each sub-block by a weighted average method to generate a fusion feature;

[0041] S502: performing enhancement processing on each high-resolution block using the fusion feature to obtain an enhanced high-resolution block;

[0042] S503: stitching all enhanced high-resolution blocks to generate a high-resolution HDR image.

[0043] The present invention also provides a high dynamic range image enhancement system based on artificial intelligence, which is applied to a high dynamic range image enhancement method based on artificial intelligence. The enhancement system includes:

[0044] A global feature extraction module, used to acquire an HDR image and extract a global feature of the HDR image, wherein the global feature is a brightness feature;

[0045] A region segmentation module, used for performing region segmentation on the HDR image based on the brightness feature to obtain a plurality of uniform initial blocks corresponding to the HDR image;

[0046] A block segmentation module, used for extracting local features of the initial block, and performing block segmentation on the initial block based on the local features to obtain a plurality of uniform sub-blocks corresponding to the initial block;

[0047] A super-resolution processing module, used to input all sub-blocks into a preset super-resolution module for super-resolution processing, and output multiple high-resolution blocks corresponding to the sub-blocks, wherein the super-resolution module is a neural network model trained by including multiple types of sample sub-blocks;

[0048] The splicing module is used to perform feature fusion and enhancement processing on the output multiple high-resolution blocks according to the global and local features of the HDR image, and to splice the processed multiple high-resolution blocks to obtain a final high-resolution HDR image.

[0049] Compared with the related art, the high dynamic range image enhancement method and enhancement system provided by the present invention based on artificial intelligence have the following beneficial effects:

[0050] The present invention can fully understand the overall brightness and local details of the image by extracting the global brightness features of the HDR image and the local features of the initial block; then, the initial block is segmented based on the local features to obtain multiple uniform sub-blocks, and the sub-blocks are super-resolved through a preset super-resolution module to output high-resolution blocks, which can retain more details in the image and improve the resolution of the image; then, the output high-resolution blocks are feature fused and enhanced, which can effectively improve the contrast and clarity of the image; finally, the processed high-resolution blocks are spliced ​​to obtain the final high-resolution HDR image, which can ensure that the final image reaches high quality in both global and local details and meet the needs of various application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A flowchart of a high dynamic range image enhancement method based on artificial intelligence provided by the present invention;

[0052] Figure 2 A module structure diagram of a high dynamic range image enhancement system based on artificial intelligence provided by the present invention. DETAILED DESCRIPTION

[0053] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only the parts related to the present invention, rather than all structures, are shown in the accompanying drawings. In addition, the embodiments of the present invention and the features in the embodiments may be combined with each other without conflict.

[0054] It should also be noted that, for ease of description, only the part relevant to the present invention but not all the content is shown in the accompanying drawings. It should be mentioned before discussing the exemplary embodiments in more detail that some exemplary embodiments are described as the processing or method described as a flow chart. Although the flow chart describes each operation (or step) as being processed sequentially, many operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of each operation can be rearranged. When its operation is completed, the processing can be terminated, but it can also have additional steps not included in the accompanying drawings. The processing can correspond to a method, a function, a procedure, a subroutine, a subprogram, etc.

[0055] Embodiment 1

[0056] The present invention provides a high dynamic range image enhancement method based on artificial intelligence, referring to Figure 1 As shown, the enhancement method includes the following steps:

[0057] S1: Acquire an HDR image, and extract a global feature of the HDR image, wherein the global feature is a brightness feature.

[0058] In this embodiment, an HDR image is first acquired, and then a brightness histogram is used to extract the global features of the HDR image regarding brightness, namely, brightness features. Through global feature extraction, the overall brightness distribution of the HDR image can be grasped, laying the foundation for subsequent regional segmentation and local feature extraction, ensuring that the entire image enhancement process can take into account global changes in brightness.

[0059] S2: performing region segmentation on the HDR image based on the brightness feature to obtain a plurality of uniform initial blocks corresponding to the HDR image.

[0060] In this embodiment, the HDR image is segmented into regions using a threshold using the extracted brightness features, and the image is divided into a plurality of initial blocks with uniform brightness. The regional segmentation helps to identify and distinguish different brightness regions in the HDR image, providing a more accurate basis for subsequent local feature extraction and refinement processing, and reducing problems such as over-smoothing or uneven brightness that may occur during global processing.

[0061] S3: extracting local features of the initial block, and performing block segmentation on the initial block based on the local features to obtain a plurality of uniform sub-blocks corresponding to the initial block.

[0062] In this embodiment, local features are further extracted from each initial block, and the local features include local brightness and edge strength. Subsequently, the initial block is subdivided again based on these local features to obtain smaller uniform sub-blocks, so as to more finely capture the subtle brightness and structural changes in the HDR image, thereby improving the accuracy and detail expression of image enhancement.

[0063] S4: input all sub-blocks into a preset super-resolution module for super-resolution processing, and output multiple high-resolution blocks corresponding to the sub-blocks, wherein the super-resolution module is a neural network model trained by containing multiple types of sample sub-blocks.

[0064] In this embodiment, all sub-blocks are sent to the super-resolution module of the pre-trained neural network for super-resolution processing. The super-resolution module is trained by a large number of sample sub-blocks of different types, and can effectively improve the resolution and clarity of the image, so as to improve the resolution of the HDR image, enhance the clarity of details, and make the image look more realistic and delicate, especially when displaying complex textures and tiny features.

[0065] S5: According to the global and local features of the HDR image, feature fusion and enhancement processing are performed on the output multiple high-resolution blocks, and the processed multiple high-resolution blocks are spliced ​​to obtain a final high-resolution HDR image.

[0066] In this embodiment, the global features and local features of the HDR image are combined through a weighted average method, and feature fusion and enhancement processing are performed on the high-resolution blocks after super-resolution processing, thereby achieving a perfect balance between global coordination and local detail enhancement of the HDR image. Not only does it retain the original dynamic range advantage of the HDR image, but it also significantly improves the visual quality and viewing experience of the image, making the image details richer and the layering stronger.

[0067] Specifically, step S1 specifically includes:

[0068] S101: Preprocessing the acquired HDR image, wherein the preprocessing includes denoising and distortion correction.

[0069] In this embodiment, after acquiring the HDR image, it is first preprocessed, including denoising and distortion correction, wherein denoising reduces image noise by applying a Gaussian filter denoising algorithm, and distortion correction eliminates lens distortion through geometric transformation to ensure the authenticity and accuracy of the image, thereby ensuring the purity and geometric accuracy of the HDR image, providing a cleaner and more accurate basis for subsequent feature extraction, and avoiding artifacts and distortion that may occur in subsequent processing.

[0070] S102: Convert the HDR image into a brightness histogram, and assign a brightness level to each column in the brightness histogram.

[0071] In this embodiment, the HDR image is converted into a brightness histogram, and each column of the histogram represents the number of pixels at a certain brightness level. This conversion can help understand the distribution of different brightness levels in the image.

[0072] S103: traverse each brightness level in the brightness histogram, and perform weighted summation of all brightness levels to obtain a brightness sum.

[0073] In this embodiment, each brightness level of the brightness histogram is traversed, and the brightness value of each level and the corresponding number of pixels are weighted summed to obtain the total brightness. The total brightness reflects the overall brightness level of the HDR image, is a key step in calculating the average brightness, and helps to comprehensively evaluate the brightness characteristics of the image.

[0074] S104: Calculate the sum of all columnar heights in the brightness histogram to obtain the total number of pixels.

[0075] In this embodiment, the sum of the heights of all columns in the brightness histogram, that is, the total number of pixels at all brightness levels, is calculated. The total number of pixels is the denominator for calculating the average brightness, thereby ensuring the accuracy of the average brightness calculation.

[0076] S105: Divide the total brightness by the total number of pixels to obtain an average brightness as a brightness feature.

[0077] In this embodiment, the total brightness is divided by the total number of pixels to obtain the average brightness as a brightness feature. The average brightness is an important component of the global features of the HDR image. It provides key information for subsequent region segmentation and local feature extraction, ensuring that the image enhancement process can fully consider the global changes in brightness, thereby achieving more accurate image processing.

[0078] Specifically, step S2 specifically includes:

[0079] S201: Determine multiple region segmentation brightness thresholds based on a statistical method.

[0080] In this embodiment, the Otsus statistical method is used to determine the brightness threshold of the HDR image. Based on the brightness histogram, the optimal threshold point is determined by analyzing the peaks and valleys in the histogram to achieve the best region segmentation effect. Specifically, the Otsus statistical method finds the optimal threshold by minimizing the intra-group variance.

[0081] S202: Compare the multiple region segmentation brightness thresholds with each pixel of the HDR image, and divide the HDR image into multiple uniform initial blocks according to the comparison result, wherein the initial block is represented as:

[0082] R i ={(x, y)|L(x, y)∈[T i-1 , T i )}

[0083] Among them, R i represents the i-th initial block, T irepresents the brightness threshold of the i-th region segmentation, [T i-1 , T i ) represents the brightness value range of the initial block; L(x, y) represents the brightness value of the position (x, y).

[0084] In this embodiment, once the region segmentation brightness threshold is determined, it is compared with the brightness value of each pixel of the HDR image. The pixels are classified into corresponding initial blocks according to the threshold interval in which the brightness value falls, thereby ensuring the consistency of brightness within each block, facilitating the subsequent accurate extraction and refined processing of local features, avoiding processing deviations caused by brightness differences, and improving the overall effect of image enhancement.

[0085] Specifically, step S3 specifically includes:

[0086] S301: traverse each of the initial blocks and extract local features of the initial blocks, wherein the local features are local brightness L l and edge strength E, respectively, are expressed as:

[0087]

[0088] Where N(x, y) represents the neighborhood around the position (x, y), I(m, n) represents the pixel value at the position (m, n), and N is the number of pixels in the neighborhood. It is expressed as the gradient strength at position (x, y).

[0089] In this embodiment, each initial block is traversed to extract its local brightness and edge strength as local features to capture more subtle brightness changes and edge information in the HDR image, providing a richer detail basis for subsequent block segmentation, and helping to retain and highlight these important features in the enhancement process.

[0090] S302: Determine the local brightness threshold T by statistical method L and edge strength threshold T E , respectively expressed as:

[0091] T L = arg min T (σ 2 (L l <T)+σ 2 (L l ≥T))

[0092] T E = arg min T (σ 2 (E <T)+σ 2 (E≥T)

[0093] Among them, T represents the dynamic candidate threshold, arg min T represents the dynamic candidate threshold T when the sum of variances is the smallest, σ 2 (L l <T) and σ 2 (L l ≥T) respectively represent the variances of the pixel values where the local luminance L l is less than the dynamic candidate threshold T and greater than or equal to the dynamic candidate threshold T, σ 2 (E<T i ) and σ 2 (E≥T i ) respectively represent the variances of the pixel values where the edge intensity E is less than the dynamic candidate threshold T and greater than or equal to the dynamic candidate threshold T.

[0094] In this embodiment, in order to perform effective block segmentation, the local luminance threshold and the edge intensity threshold are determined by statistical methods, which can more accurately distinguish different regions in the HDR image, ensuring that the block segmentation takes into account both the luminance uniformity and the integrity of edge details, thus avoiding the problems of over-smoothing and detail loss in subsequent image enhancement processing.

[0095] S303: Segment each initial block based on the local luminance threshold T L and the edge intensity threshold T E to obtain sub-blocks, where the sub-blocks are represented as:

[0096] C j ={(x, y)∈B|L l ∈[T L(j-1) , T L(j) ) and E>T E}

[0097] where C j represents the j-th sub-block, [T L(j-1) , T L(j) ) represents the range of local luminance values of the j-th sub-block, and B represents the initial block.

[0098] In this embodiment, each sub-block C j represents a set of pixels with luminance values within a specific range. Through block segmentation based on local luminance and edge intensity thresholds, the granularity of image enhancement can be more finely controlled, ensuring that the luminance and edge information within each sub-block are relatively consistent, thus achieving better visual effects and detail retention in subsequent super-resolution processing and feature fusion.

[0099] Specifically, step S4 specifically includes:

[0100] S401: Pass each sub-block as input to a pre-trained neural network-based super-resolution model.

[0101] In this embodiment, all the sub-blocks that have been preliminarily processed are sent one by one to a pre-trained neural network model, which is specially designed for super-resolution processing, that is, improving the resolution of the image. The architecture of the model includes a convolutional neural network (CNN), which is trained with a large amount of sample data and can effectively restore high-resolution details from low-resolution images. By passing the sub-blocks as input to the super-resolution model, the deep learning ability of the model can be used to intelligently infer and reconstruct the missing high-frequency details, thereby significantly improving the clarity and sharpness of the image.

[0102] S402: Perform super-resolution processing on each input sub-block using the super-resolution model, and improve the resolution of each sub-block through the learned weight and bias parameters.

[0103] In this embodiment, within the model, each sub-block is processed through a series of convolutional layers, upsampling layers, activation functions, and jump connections to gradually increase the resolution of the image and restore details. The model maps low-resolution input to high-resolution output through learned weights and bias parameters. Each sub-block can obtain significantly enhanced details and clarity while maintaining the original content.

[0104] S403: The super-resolution model outputs a high-resolution block corresponding to each input sub-block.

[0105] In this embodiment, when the super-resolution model completes processing of each sub-block, it will output the corresponding high-resolution version of the block. These high-resolution blocks retain the content of the atomic block but have a higher pixel density and richer details. The output high-resolution blocks will then be used for the next step of feature fusion and enhancement processing. Through the processing of the super-resolution model, the originally low-resolution sub-blocks are converted into high-resolution blocks, greatly improving the detail expression and visual effect of the image.

[0106] Specifically, step S5 specifically includes:

[0107] S501: Combine the global feature and the local feature of each sub-block by a weighted average method to generate a fusion feature.

[0108] In this embodiment, a weighted average strategy is adopted to combine the previously extracted global brightness features of the HDR image with the local features of each sub-block, and a pre-trained deep convolutional neural network is used to forward propagate the entire HDR image to extract the global feature vector; the HDR image is divided into multiple high-resolution sub-blocks, and a smaller convolution kernel is applied to each sub-block for convolution operation to extract local features; for each sub-block, a weight coefficient matrix is ​​defined, in which the elements represent the relative importance of the global features and the local features, and finally the two parts of the features are fused together through a learnable weighted layer.

[0109] The weight coefficient matrix may be a preset fixed value or may be dynamically calculated through an additional learning process.

[0110] S502: Perform enhancement processing on each high-resolution block using the fusion feature to obtain an enhanced high-resolution block.

[0111] In this embodiment, each high-resolution block is further enhanced based on the fusion features, including but not limited to adjusting contrast, sharpening edges, enhancing textures, etc., to enhance the visual impact and detail expression of the image, which can significantly improve the visual quality of the image, so that it can show more distinct details and layering while maintaining the HDR characteristics.

[0112] S503: stitching all enhanced high-resolution blocks to generate a high-resolution HDR image.

[0113] In this embodiment, all enhanced high-resolution blocks are spliced ​​according to their original positions to reconstruct a complete high-resolution HDR image, which not only significantly improves the resolution, but also optimizes the details, contrast and color performance. This image can better show the dynamic range advantage of HDR images, while providing a more delicate and vivid visual experience, and is suitable for various high-demand visual application scenarios, such as high-end displays, professional image processing and virtual reality environments.

[0114] The working principle of the high dynamic range image enhancement method based on artificial intelligence provided by the present invention is as follows: by extracting the global brightness features of the HDR image and the local features of the initial block, the overall brightness and local details of the image can be fully understood; then, the initial block is segmented based on the local features to obtain multiple uniform sub-blocks, and the sub-blocks are super-resolved through a preset super-resolution module to output high-resolution blocks, which can retain more details in the image and improve the resolution of the image at the same time; then, the output high-resolution blocks are feature fused and enhanced, which can effectively improve the contrast and clarity of the image; finally, the processed high-resolution blocks are spliced ​​to obtain the final high-resolution HDR image, which can ensure that the final image reaches a high quality in both global and local details and meet the needs of various application scenarios.

[0115] Embodiment 2

[0116] The present invention also provides a high dynamic range image enhancement system based on artificial intelligence, which is applied to a high dynamic range image enhancement system based on artificial intelligence, referring to Figure 2 As shown, the enhanced system includes:

[0117] The global feature extraction module 100 is used to acquire an HDR image and extract a global feature of the HDR image, wherein the global feature is a brightness feature.

[0118] The region segmentation module 200 is used to perform region segmentation on the HDR image based on the brightness feature to obtain a plurality of uniform initial blocks corresponding to the HDR image.

[0119] The block segmentation module 300 is used to extract local features of the initial block and perform block segmentation on the initial block based on the local features to obtain a plurality of uniform sub-blocks corresponding to the initial block.

[0120] The super-resolution processing module 400 is used to input all sub-blocks into a preset super-resolution module for super-resolution processing, and output multiple high-resolution blocks corresponding to the sub-blocks, wherein the super-resolution module is a neural network model trained by containing multiple types of sample sub-blocks.

[0121] The stitching module 500 is used to perform feature fusion and enhancement processing on the output multiple high-resolution blocks according to the global and local features of the HDR image, and stitch the processed multiple high-resolution blocks to obtain a final high-resolution HDR image.

[0122] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0123] A person skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0124] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or includes elements that are inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device that includes the elements.

Claims

1. A high dynamic range image enhancement method based on artificial intelligence, characterized in that: The enhancement method includes the following steps: S1: Acquire an HDR image, and extract a global feature of the HDR image, wherein the global feature is a brightness feature; S2: performing region segmentation on the HDR image based on the brightness feature to obtain a plurality of uniform initial blocks corresponding to the HDR image; S3: extracting local features of the initial block, and performing block segmentation on the initial block based on the local features to obtain a plurality of uniform sub-blocks corresponding to the initial block; Wherein step S3 specifically comprises: S301: traverse each of the initial blocks and extract local features of the initial blocks, wherein the local features are local brightness and edge strength , respectively expressed as: in, Represented as location The surrounding neighborhood, Represented as location The pixel value of is the number of pixels in the neighborhood, Represented as location The gradient strength; S302: Determine a local brightness threshold by using a statistical method and edge strength threshold , respectively expressed as: in, represents the dynamic candidate threshold, Expressed as the dynamic candidate threshold with the minimum sum of variances , and Respectively represent the local brightness Less than the dynamic candidate threshold The sum is greater than or equal to the dynamic candidate threshold The variance of the pixel values, and The edge strength Less than the dynamic candidate threshold The sum is greater than or equal to the dynamic candidate threshold The variance of the pixel values; S303: Based on the local brightness threshold and edge strength threshold Each initial block is divided into sub-blocks, where the sub-blocks are represented as: in, Expressed as Sub-blocks, Indicates The local brightness value range of sub-blocks, Represented as initial block; S4: inputting all sub-blocks into a preset super-resolution module for super-resolution processing, and outputting a plurality of high-resolution blocks corresponding to the sub-blocks, wherein the super-resolution module is a neural network model trained by including multiple types of sample sub-blocks; S5: According to the global and local features of the HDR image, feature fusion and enhancement processing are performed on the output multiple high-resolution blocks, and the processed multiple high-resolution blocks are spliced ​​to obtain a final high-resolution HDR image.

2. The high dynamic range image enhancement method based on artificial intelligence according to claim 1, characterized in that: Step S1 specifically includes: S101: Preprocessing the acquired HDR image, wherein the preprocessing includes denoising and distortion correction; S102: converting the HDR image into a brightness histogram, and assigning a brightness level to each column in the brightness histogram; S103: traversing each brightness level in the brightness histogram, and weighted summing all brightness levels to obtain a brightness sum; S104: Calculate the sum of all columnar heights in the brightness histogram to obtain the total number of pixels; S105: Divide the total brightness by the total number of pixels to obtain an average brightness as a brightness feature.

3. The high dynamic range image enhancement method based on artificial intelligence according to claim 2, characterized in that: Step S2 specifically includes: S201: Determine multiple region segmentation brightness thresholds based on a statistical method; S202: Compare the multiple region segmentation brightness thresholds with each pixel of the HDR image, and divide the HDR image into multiple uniform initial blocks according to the comparison result, wherein the initial block is represented as: in, Indicates Initial blocks, Indicates Region segmentation brightness threshold, Represented as the brightness value range of the initial block; Represented as location The brightness value of the 4. The high dynamic range image enhancement method based on artificial intelligence according to claim 3, characterized in that: Step S4 specifically includes: S401: passing each sub-block as input to a pre-trained neural network-based super-resolution model; S402: performing super-resolution processing on each input sub-block using the super-resolution model, and improving the resolution of each sub-block through the learned weight and bias parameters; S403: The super-resolution model outputs a high-resolution block corresponding to each input sub-block.

5. The high dynamic range image enhancement method based on artificial intelligence according to claim 4, characterized in that: Step S5 specifically includes: S501: combining the global feature and the local feature of each sub-block by a weighted average method to generate a fusion feature; S502: performing enhancement processing on each high-resolution block using the fusion feature to obtain an enhanced high-resolution block; S503: stitching all enhanced high-resolution blocks to generate a high-resolution HDR image.

6. A high dynamic range image enhancement system based on artificial intelligence, applied to a high dynamic range image enhancement method based on artificial intelligence as claimed in any one of claims 1 to 5, characterized in that: Enhanced systems include: A global feature extraction module, used to acquire an HDR image and extract a global feature of the HDR image, wherein the global feature is a brightness feature; A region segmentation module, used for performing region segmentation on the HDR image based on the brightness feature to obtain a plurality of uniform initial blocks corresponding to the HDR image; A block segmentation module, used for extracting local features of the initial block, and performing block segmentation on the initial block based on the local features to obtain a plurality of uniform sub-blocks corresponding to the initial block; The block segmentation module is also used to perform the following steps: First, traverse each of the initial blocks and extract the local features of the initial blocks, wherein the local features are local brightness and edge strength , respectively expressed as: in, Represented as location The surrounding neighborhood, Represented as location The pixel value of is the number of pixels in the neighborhood, Represented as location The gradient strength; Secondly, the local brightness threshold is determined by statistical methods and edge strength threshold , respectively expressed as: in, represents the dynamic candidate threshold, Expressed as the dynamic candidate threshold with the minimum sum of variances , and Respectively represent the local brightness Less than the dynamic candidate threshold The sum is greater than or equal to the dynamic candidate threshold The variance of the pixel values, and The edge strength Less than the dynamic candidate threshold The sum is greater than or equal to the dynamic candidate threshold The variance of the pixel values; Finally, based on the local brightness threshold and edge strength threshold Each initial block is divided into sub-blocks, where the sub-blocks are represented as: in, Expressed as Sub-blocks, Indicates The local brightness value range of sub-blocks, Represented as initial block; A super-resolution processing module, used to input all sub-blocks into a preset super-resolution module for super-resolution processing, and output multiple high-resolution blocks corresponding to the sub-blocks, wherein the super-resolution module is a neural network model trained by including multiple types of sample sub-blocks; The splicing module is used to perform feature fusion and enhancement processing on the output multiple high-resolution blocks according to the global and local features of the HDR image, and to splice the processed multiple high-resolution blocks to obtain a final high-resolution HDR image.

Images