An intelligent photography image data processing system based on AI scene recognition

Abstract

This invention provides an intelligent photographic image data processing system based on AI scene recognition, relating to the field of image data technology. It includes an image acquisition module for acquiring photographic image data to be processed, normalizing the photographic image data to be processed, and dividing the normalized photographic image data to be processed into multiple image units to obtain the image to be processed. This intelligent photographic image data processing system based on AI scene recognition can constrain the risks of local over-enhancement, noise amplification, texture loss, noise edge amplification, and color shift that may be caused by the combined effect of multiple image processing parameters before performing image enhancement and restoration processing. This allows the final target image data to achieve a more stable balance between brightness level, detail clarity, noise suppression, texture preservation, and color naturalness, improving the refinement and visual quality of image data processing results in complex photographic scenes.

Description

Technical Field

[0001] This invention relates to the field of image data technology, specifically to an intelligent photographic image data processing system based on AI scene recognition. Background Technology

[0002] In existing technologies, intelligent photographic image data processing typically involves acquiring the photographic image and then performing processes such as brightness adjustment, contrast enhancement, noise reduction, sharpening, color correction, and dynamic range adjustment to improve the overall visual effect of the image. Some solutions also incorporate artificial intelligence scene recognition to identify scene features in the photographic image, such as night scenes, portraits, landscapes, indoor scenes, and backlighting, and select appropriate image processing templates or parameters based on the recognition results, thereby automating the enhancement process of the photographic image.

[0003] However, existing technologies typically only use scene recognition results as the basis for global processing of the entire image. It is difficult to generate adaptive image processing parameters at the region or pixel level based on scene features, brightness distribution, texture details, and noise levels in different regions of the image. This results in the same or similar processing strategies being applied to bright areas, dark areas, subject areas, and background areas in the same image. This can easily lead to problems such as local over-enhancement, insufficient preservation of dark details, noise amplification, texture loss, or color distortion. It is difficult to meet the needs of fine-grained enhancement and restoration of image data in complex photographic scenes. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent photographic image data processing system based on AI scene recognition. By generating adaptive processing parameters at the image unit level through scene sensitivity maps and multi-parameter coupling risk correction, it solves the problems of inaccurate local image enhancement, noise amplification, texture loss, and color distortion in complex scenes.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent photographic image data processing system based on AI scene recognition, comprising:

[0006] The image acquisition module is used to acquire photographic image data to be processed, normalize the photographic image data to be processed, and divide the normalized photographic image data to be processed into multiple image units to obtain the image to be processed.

[0007] The scene feature analysis module is used to perform AI scene recognition on the image to be processed, generate a scene region confidence map, and generate a scene sensitivity map based on the scene region confidence map. The scene sensitivity map includes a processing sensitivity vector corresponding to each image unit. The processing sensitivity vector is used to characterize the sensitivity of the corresponding image unit to brightness adjustment, contrast enhancement, noise reduction, sharpening, and color correction processing. The scene region confidence map is used to characterize the correlation between each image unit and the corresponding photographic scene. The scene sensitivity map is used to characterize the sensitivity of each image unit to brightness adjustment, contrast enhancement, noise reduction, sharpening, and color correction processing.

[0008] The local image quality analysis module is used to extract the brightness state, texture state and noise state of each image unit and generate a local image quality map corresponding to the scene sensitivity map space. The local image quality map includes a brightness distribution map, a texture preservation weight map and a noise estimation map.

[0009] An adaptive parameter generation module is used to generate an initial image processing parameter map based on the scene sensitivity map and the local image quality map; and based on the processing sensitivity vector, to weight the parameter conflict relationships among brightness adjustment parameters, contrast enhancement parameters, noise reduction parameters, sharpening parameters, and color correction parameters in the initial image processing parameter map, constructing a parameter conflict response matrix corresponding to each image unit; to generate a multi-parameter coupling risk map based on the parameter conflict response matrix, the multi-parameter coupling risk map being used to characterize the local over-enhancement risk, noise amplification risk, texture loss risk, noise edge amplification risk, and color shift risk caused by multiple image processing parameters in each image unit; and to perform reverse constraint correction on the initial image processing parameter map based on the multi-parameter coupling risk map to obtain the target image processing parameter map.

[0010] The image processing output module is used to perform adaptive image enhancement and restoration processing on each image unit in the image to be processed according to the target image processing parameter map, so as to obtain target image data.

[0011] Preferably, when the image acquisition module performs normalization processing on the photographic image data to be processed, it performs color space unification, pixel value range mapping and image size standardization processing, and divides the normalized photographic image data to be processed into multiple image units according to a preset pixel size. An overlapping boundary band is set between adjacent image units. The overlapping boundary band is used for the image processing output module to perform transition fusion on the processing results of adjacent image units.

[0012] Preferably, the scene region confidence map includes a scene confidence vector corresponding to each image unit, and the scene confidence vector includes the confidence values ​​of the corresponding image unit belonging to the highlight region, dark region, subject region, background region and high texture region, respectively.

[0013] Preferably, the scene feature analysis module generates a processing sensitivity vector corresponding to each image unit based on the scene confidence vector and a preset scene processing sensitivity mapping relationship. The processing sensitivity vector includes brightness adjustment sensitivity, contrast enhancement sensitivity, noise reduction sensitivity, sharpening sensitivity, and color correction sensitivity.

[0014] Preferably, when generating the local image quality map, the local image quality analysis module calculates the average brightness value, the proportion of bright pixels and the proportion of dark pixels of each image unit to generate the brightness distribution map, calculates the gradient intensity and edge density of each image unit to generate the texture preservation weight map, and generates the noise estimation map based on the pixel fluctuation amplitude of the flat sub-region within each image unit.

[0015] Preferably, the adaptive parameter generation module combines the processing sensitivity vector, brightness distribution value, texture preservation weight value and noise estimation value corresponding to each image unit to generate an initial processing parameter vector corresponding to that image unit. The initial processing parameter vector includes brightness adjustment parameters, contrast enhancement parameters, noise reduction parameters, sharpening parameters and color correction parameters. The initial processing parameter vectors corresponding to each image unit together constitute the initial image processing parameter map.

[0016] Preferably, the parameter conflict response matrix is ​​a matrix that corresponds one-to-one with each image unit. The rows and columns of the parameter conflict response matrix correspond to the brightness adjustment parameter, contrast enhancement parameter, noise reduction parameter, sharpening parameter, and color correction parameter, respectively. The matrix elements in the parameter conflict response matrix are used to characterize the influence intensity of the processing risk caused by different image processing parameters.

[0017] Preferably, when the adaptive parameter generation module constructs the parameter conflict response matrix, it uses the processing sensitivity vector corresponding to each image unit as a weighting factor to perform weighting processing on the preset parameter conflict basis matrix to obtain the parameter conflict response matrix of the corresponding image unit. The parameter conflict basis matrix includes the conflict relationship between brightness adjustment parameters and noise amplification risk, the conflict relationship between contrast enhancement parameters and local over-enhancement risk, the conflict relationship between noise reduction parameters and texture loss risk, the conflict relationship between sharpening parameters and noise edge amplification risk, and the conflict relationship between color correction parameters and color shift risk.

[0018] Preferably, the adaptive parameter generation module generates a risk vector for each image unit based on the initial processing parameter vector and parameter conflict response matrix corresponding to each image unit. The risk vector includes a local over-enhancement risk component, a noise amplification risk component, a texture loss risk component, a noise edge amplification risk component, and a color shift risk component. The risk vectors corresponding to each image unit together constitute the multi-parameter coupled risk map.

[0019] Preferably, when performing reverse constraint correction, the adaptive parameter generation module compares the risk vector of each image unit with a preset risk threshold vector to obtain a risk exceeding the limit vector, and generates a parameter correction coefficient vector based on the risk exceeding the limit vector; it then corrects the initial processing parameter vector of the image unit based on the parameter correction coefficient vector to obtain the target processing parameter vector of the image unit, and the target processing parameter vectors of each image unit together constitute the target image processing parameter map.

[0020] This invention provides an intelligent photographic image data processing system based on AI scene recognition. It has the following beneficial effects:

[0021] This AI-based intelligent photographic image data processing system divides the photographic image to be processed into multiple image units and generates scene region confidence maps and scene sensitivity maps based on the AI ​​scene recognition results. This transforms the scene recognition results from simply serving as the global processing basis for the entire image into processing sensitivity vectors corresponding to each image unit. Simultaneously, it generates an initial image processing parameter map by combining a brightness distribution map, a texture preservation weight map, and a noise estimation map. This allows for the determination of differentiated image processing requirements for bright areas, dark areas, subject areas, background areas, and high-texture areas within the same photographic image. This improves the matching degree between image enhancement and restoration processing and specific photographic scenes, avoiding problems such as local over-enhancement, insufficient preservation of dark details, noise amplification, texture loss, and color distortion caused by using uniform processing parameters.

[0022] Furthermore, after generating the initial image processing parameter map, this invention weights the parameter conflict relationships among brightness adjustment parameters, contrast enhancement parameters, noise reduction parameters, sharpening parameters, and color correction parameters based on the processing sensitivity vector, constructs a parameter conflict response matrix corresponding to each image unit, and generates a multi-parameter coupling risk map accordingly. Then, based on the risk exceeding limits, the initial processing parameter vector is reverse-constrained and corrected to obtain the target image processing parameter map. Thus, this invention can constrain the risks of local over-enhancement, noise amplification, texture loss, noise edge amplification, and color shift that may be caused by the combined effect of multiple image processing parameters before performing image enhancement and restoration processing. This allows the final target image data to achieve a more stable balance between brightness levels, detail clarity, noise suppression, texture preservation, and color naturalness, improving the refinement and visual quality of image data processing results in complex photographic scenes. Attached Figure Description

[0023] Figure 1 This is a system structure diagram of the present invention;

[0024] Figure 2 This is a flowchart illustrating the generation process of the scene region confidence map and scene sensitivity map in this invention.

[0025] Figure 3 This is a flowchart illustrating the generation process of the local image quality map and the initial image processing parameter map in this invention.

[0026] Figure 4 This is a flowchart of the multi-parameter coupling risk graph generation and reverse constraint correction process of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Example 1

[0029] like Figure 1 As shown, this embodiment of the invention provides an intelligent photographic image data processing system based on AI scene recognition, including an image acquisition module, used to acquire photographic image data to be processed, perform normalization processing on the photographic image data to be processed, and divide the normalized photographic image data to be processed into multiple image units to obtain the image to be processed.

[0030] When the image acquisition module performs normalization processing on the photographic image data to be processed, it performs color space unification, pixel value range mapping and image size standardization processing. It divides the normalized photographic image data to be processed into multiple image units according to the preset pixel size. An overlapping boundary band is set between adjacent image units. The overlapping boundary band is used for the image processing output module to transition and fuse the processing results of adjacent image units.

[0031] The scene feature analysis module is used to perform AI scene recognition on the image to be processed, generate a scene region confidence map, and generate a scene sensitivity map based on the scene region confidence map. The scene sensitivity map includes a processing sensitivity vector corresponding to each image unit. The processing sensitivity vector is used to characterize the sensitivity of the corresponding image unit to brightness adjustment, contrast enhancement, noise reduction, sharpening, and color correction processing. The scene region confidence map is used to characterize the correlation between each image unit and the corresponding photographic scene. The scene sensitivity map is used to characterize the sensitivity of each image unit to brightness adjustment, contrast enhancement, noise reduction, sharpening, and color correction processing.

[0032] The scene region confidence map includes scene confidence vectors corresponding to each image unit. These vectors contain confidence values ​​for each image unit belonging to a highlight region, dark region, subject region, background region, and high-texture region, respectively. The scene feature analysis module generates processing sensitivity vectors for each image unit based on the scene confidence vectors and a pre-defined scene processing sensitivity mapping relationship. These processing sensitivity vectors include sensitivity for brightness adjustment, contrast enhancement, noise reduction, sharpening, and color correction.

[0033] The local image quality analysis module is used to extract the brightness state, texture state, and noise state of each image unit and generate a local image quality map corresponding to the scene sensitivity map space. The local image quality map includes a brightness distribution map, a texture preservation weight map, and a noise estimation map.

[0034] When generating a local image quality map, the local image quality analysis module calculates the average brightness value, the proportion of bright pixels and the proportion of dark pixels in each image unit to generate a brightness distribution map, calculates the gradient intensity and edge density of each image unit to generate a texture preservation weight map, and generates a noise estimation map based on the pixel fluctuation amplitude of flat sub-regions within each image unit.

[0035] The adaptive parameter generation module generates an initial image processing parameter map based on the scene sensitivity map and the local image quality map. Based on the processing sensitivity vector, it weights the parameter conflict relationships among brightness adjustment parameters, contrast enhancement parameters, noise reduction parameters, sharpening parameters, and color correction parameters in the initial image processing parameter map, constructs a parameter conflict response matrix for each image unit, and generates a multi-parameter coupling risk map based on the parameter conflict response matrix. The multi-parameter coupling risk map is used to characterize the risks of local over-enhancement, noise amplification, texture loss, noise edge amplification, and color shift caused by multiple image processing parameters in each image unit. The initial image processing parameter map is then subjected to inverse constraint correction based on the multi-parameter coupling risk map to obtain the target image processing parameter map.

[0036] For each image unit, the adaptive parameter generation module combines the processing sensitivity vector, brightness distribution value, texture preservation weight value, and noise estimation value corresponding to the image unit to generate the initial processing parameter vector corresponding to that image unit. The initial processing parameter vector includes brightness adjustment parameters, contrast enhancement parameters, noise reduction parameters, sharpening parameters, and color correction parameters. The initial processing parameter vectors corresponding to each image unit together constitute the initial image processing parameter map.

[0037] The parameter conflict response matrix is ​​a matrix that corresponds one-to-one with each image unit. The rows and columns of the parameter conflict response matrix correspond to the brightness adjustment parameter, contrast enhancement parameter, noise reduction parameter, sharpening parameter, and color correction parameter, respectively. The matrix elements in the parameter conflict response matrix are used to characterize the intensity of the impact of different image processing parameters on the processing risk.

[0038] When constructing the parameter conflict response matrix, the adaptive parameter generation module uses the processing sensitivity vector corresponding to each image unit as a weighting factor to perform weighting processing on the preset parameter conflict basis matrix, thereby obtaining the parameter conflict response matrix of the corresponding image unit. The parameter conflict basis matrix includes the conflict relationship between brightness adjustment parameters and noise amplification risk, the conflict relationship between contrast enhancement parameters and local over-enhancement risk, the conflict relationship between noise reduction parameters and texture loss risk, the conflict relationship between sharpening parameters and noise edge amplification risk, and the conflict relationship between color correction parameters and color shift risk.

[0039] The adaptive parameter generation module generates a risk vector for each image unit based on the initial processing parameter vector and parameter conflict response matrix. The risk vector includes local over-enhancement risk components, noise amplification risk components, texture loss risk components, noise edge amplification risk components, and color shift risk components. The risk vectors for each image unit together constitute a multi-parameter coupled risk map.

[0040] When performing reverse constraint correction, the adaptive parameter generation module compares the risk vector of each image unit with the preset risk threshold vector to obtain the risk exceeding the limit vector, and generates the parameter correction coefficient vector based on the risk exceeding the limit vector; it then corrects the initial processing parameter vector of the image unit based on the parameter correction coefficient vector to obtain the target processing parameter vector of the image unit. The target processing parameter vectors of each image unit together constitute the target image processing parameter map.

[0041] The image processing output module is used to perform adaptive image enhancement and restoration processing on each image unit in the image to be processed according to the target image processing parameter map, so as to obtain the target image data.

[0042] Example 2

[0043] like Figure 2 As shown in Example 1, this example provides a detailed explanation of the process by which the scene feature analysis module generates scene region confidence maps and scene sensitivity maps.

[0044] The scene feature analysis module takes each image unit as the processing object, performs AI scene recognition on the image to be processed, and obtains the scene confidence vector corresponding to each image unit. For the first image unit... Image unit Its scenario confidence vector can be represented as:

[0045]

[0046] in, These represent the confidence values ​​for the image unit to belong to the highlight region, dark region, subject region, background region, and high-texture region, respectively. The scene confidence vectors corresponding to multiple image units are arranged according to their spatial positions in the image to be processed, forming a scene region confidence map.

[0047] In this embodiment, the scene feature analysis module pre-sets a scene processing sensitivity mapping relationship, which represents the sensitivity of different photographic scene regions to different image processing parameters. For example, the scene processing sensitivity mapping relationship can be represented as the following mapping table:

[0048] Table 1: Scene processing sensitivity mapping table.

[0049] Photography Scene Area Brightness adjustment sensitivity Contrast Enhancement Sensitivity Noise reduction sensitivity Sharpening sensitivity Color correction sensitivity Highlighted area 0.90 0.85 0.30 0.40 0.60 Dark areas 0.85 0.65 0.90 0.55 0.50 Main area 0.60 0.50 0.45 0.80 0.85 Background area 0.45 0.40 0.75 0.35 0.45 High texture area 0.50 0.70 0.55 0.90 0.40

[0050] The values ​​above illustrate the sensitivity of different areas in a photographic scene to different image processing operations. For example, dark areas are more sensitive to brightness adjustment and noise reduction, high-texture areas are more sensitive to sharpening and texture preservation, and the main subject area is more sensitive to sharpening and color correction.

[0051] For the Image unit The scene feature analysis module analyzes the scene confidence vector of the image unit. The image unit is mapped to the scene processing sensitivity, and a processing sensitivity vector is generated accordingly.

[0052]

[0053] in, These represent the sensitivity of the image unit to brightness adjustment, contrast enhancement, noise reduction, sharpening, and color correction, respectively. The processing sensitivity vectors corresponding to multiple image units are arranged according to their spatial location to form a scene sensitivity map.

[0054] For example, for image units The scene confidence vector obtained through AI scene recognition is as follows:

[0055]

[0056] This indicates that the image unit mainly possesses features of dark and background regions, while also containing some features of high-texture and subject regions. Based on the aforementioned scene processing sensitivity mapping relationship, the processing sensitivity corresponding to each scene region is weighted and calculated to obtain the processing sensitivity vector corresponding to this image unit:

[0057]

[0058] Here, 0.65 indicates that the image unit has high sensitivity to brightness adjustment, 0.73 indicates that the image unit has high sensitivity to noise reduction, and 0.55 indicates that the image unit has medium sensitivity to sharpening. Therefore, when generating subsequent image processing parameters, this image unit should focus on the relationship between enhancing dark area brightness and suppressing noise, avoiding noise amplification caused by brightness adjustment.

[0059] In this embodiment, the AI ​​scene recognition results are no longer simply output as scene category labels, but are converted into processing sensitivity vectors corresponding to each image unit, and further constitute a scene sensitivity map. The scene sensitivity map serves as the data foundation for subsequently generating initial image processing parameter maps, parameter conflict response matrices, and multi-parameter coupling risk maps, enabling subsequent image enhancement and restoration processing to match the processing needs of different photographic scene regions.

[0060] Example 3

[0061] like Figure 3As shown, based on Embodiments 1 and 2, this embodiment describes the process of the local image quality analysis module generating a local image quality map and the adaptive parameter generation module generating an initial image processing parameter map.

[0062] The local image quality analysis module takes each image unit as the processing object and extracts the brightness state, texture state, and noise state of each image unit. For any image unit, the local image quality analysis module obtains the brightness channel data of that image unit and calculates the average brightness value, the proportion of bright pixels, and the proportion of dark pixels of that image unit to generate a brightness distribution value. The brightness distribution values ​​corresponding to multiple image units are arranged according to their spatial position in the image to be processed to form a brightness distribution map.

[0063] When extracting texture state, the local image quality analysis module calculates the gradient intensity and edge density of each image unit to generate texture preservation weight values. The texture preservation weight values ​​corresponding to multiple image units are arranged according to their spatial location to form a texture preservation weight map. Gradient intensity characterizes the magnitude of pixel variation within an image unit, while edge density characterizes the distribution ratio of edge pixels within that unit. Therefore, when an image unit contains strong texture or abundant edge information, its corresponding texture preservation weight value is larger, reducing texture loss caused by excessive noise reduction or smoothing during subsequent image processing.

[0064] When extracting noise states, the local image quality analysis module determines flat sub-regions in each image unit and generates noise estimates based on the pixel fluctuation amplitude within the flat sub-regions. The noise estimates corresponding to multiple image units are arranged according to their spatial positions to form a noise estimation map. The flat sub-regions are those where the gradient intensity is lower than a preset gradient threshold, and the pixel fluctuation amplitude can be determined by the standard deviation or mean absolute deviation of the pixel values ​​within the flat sub-regions.

[0065] After generating the local image quality map, the adaptive parameter generation module combines the processing sensitivity vector, brightness distribution value, texture preservation weight value, and noise estimate value corresponding to each image unit to generate the initial processing parameter vector for that image unit. The initial processing parameter vector includes brightness adjustment parameters, contrast enhancement parameters, noise reduction parameters, sharpening parameters, and color correction parameters. Among them, the brightness adjustment parameter is jointly determined by the brightness adjustment sensitivity and brightness distribution value; the contrast enhancement parameter is jointly determined by the contrast enhancement sensitivity and texture preservation weight value; the noise reduction parameter is jointly determined by the noise reduction sensitivity and noise estimate value; the sharpening parameter is jointly determined by the sharpening sensitivity, texture preservation weight value, and noise estimate value; and the color correction parameter is jointly determined by the color correction sensitivity and brightness state.

[0066] Specifically, for image units Its processing sensitivity vector is:

[0067]

[0068] The local image quality analysis module obtained a brightness distribution value of 0.53, a texture preservation weight value of 0.43, and a noise estimate value of 0.18 for this image unit. The adaptive parameter generation module combines these values ​​to obtain the final image unit. Initial processing parameter vector:

[0069]

[0070] Wherein, 0.59 represents the brightness adjustment parameter, 0.51 represents the contrast enhancement parameter, 0.54 represents the noise reduction parameter, 0.45 represents the sharpening parameter, and 0.43 represents the color correction parameter. The initial processing parameter vectors corresponding to multiple image units are arranged according to their spatial positions in the image to be processed, and together they constitute the initial image processing parameter map.

[0071] In this embodiment, the scene sensitivity map is combined with the local image quality map in the above manner, so that the initial image processing parameters are no longer determined solely by the scene category or the global image quality, but are jointly determined by the scene processing sensitivity, brightness state, texture state and noise state of each image unit. This provides a data foundation for the subsequent construction of the parameter conflict response matrix, the multi-parameter coupling risk map and the target image processing parameter map.

[0072] Example 4

[0073] like Figure 4 As shown, based on Examples 1 to 3, this example describes the process of the adaptive parameter generation module constructing the parameter conflict response matrix, generating a multi-parameter coupled risk graph, and performing reverse constraint correction.

[0074] The adaptive parameter generation module constructs a corresponding parameter conflict response matrix for each image unit. For the ... Image unit Its initial processing parameter vector is represented as:

[0075]

[0076] in, These represent the brightness adjustment parameters, contrast enhancement parameters, noise reduction parameters, sharpening parameters, and color correction parameters, respectively.

[0077] The adaptive parameter generation module pre-sets a parameter conflict basis matrix, which characterizes the fundamental influence strength of different image processing parameters on processing risks. Unlike directly generating risk values ​​from the initial processing parameter vector, this embodiment first weights the parameter conflict basis matrix using processing sensitivity vectors, ensuring that the same image processing parameters produce different risk responses in different photographic scenarios. Specifically, dark areas are more sensitive to noise amplification caused by brightness adjustments, main subject areas are more sensitive to color shifts caused by color correction, and high-texture areas are more sensitive to texture loss caused by noise reduction. Therefore, the parameter conflict response matrix is ​​not a fixed matrix but is generated by the processing sensitivity vectors of each image unit.

[0078] In one implementation, the parameter conflict basis matrix includes the following main conflict terms: the conflict relationship between brightness adjustment parameters and noise amplification risk; the conflict relationship between contrast enhancement parameters and local over-enhancement risk; the conflict relationship between noise reduction parameters and texture loss risk; the conflict relationship between sharpening parameters and noise edge amplification risk; and the conflict relationship between color correction parameters and color shift risk. The adaptive parameter generation module uses the processing sensitivity vector corresponding to each image unit as a weighting factor to weight the main conflict terms, obtaining the parameter conflict response intensity corresponding to that image unit. Then, it generates a risk vector based on the parameter conflict response intensity and the initial processing parameter vector.

[0079] To illustrate the unconventional calculation process described above, the following uses image units as an example. For example, as can be seen from Embodiments 2 and 3, the image unit... The processing sensitivity vector and the initial processing parameter vector are as follows:

[0080]

[0081] in, The five values ​​correspond to the sensitivity of brightness adjustment, contrast enhancement, noise reduction, sharpening, and color correction, respectively. The five values ​​correspond to the brightness adjustment parameter, contrast enhancement parameter, noise reduction parameter, sharpening parameter, and color correction parameter, respectively.

[0082] The parameter conflict base strength vector is set as follows:

[0083]

[0084] Among them, 0.70 corresponds to the basic strength of the risk of noise amplification caused by the brightness adjustment parameter, 0.65 corresponds to the basic strength of the risk of local over-enhancement caused by the contrast enhancement parameter, 0.60 corresponds to the basic strength of the risk of texture loss caused by the noise reduction parameter, 0.75 corresponds to the basic strength of the risk of noise edge amplification caused by the sharpening parameter, and 0.55 corresponds to the basic strength of the risk of color shift caused by the color correction parameter.

[0085] The adaptive parameter generation module generates parameters based on the processing sensitivity vector. For parameter conflict basic strength vector By weighting, we obtain the parameter conflict response intensity vector:

[0086]

[0087] Where ⨀ represents multiplication by corresponding positions. This calculation indicates that the same set of basic conflict relationships exists in the image unit. The conflict response intensity is adjusted by the scene processing sensitivity of the image unit, thereby obtaining a conflict response intensity that is adapted to the scene features of the image unit.

[0088] Subsequently, the adaptive parameter generation module generates parameters based on the parameter conflict response intensity vector. and initial processing parameter vector Generate a risk vector:

[0089]

[0090] in, The five risk components correspond to the noise amplification risk component, the local over-enhancement risk component, the texture loss risk component, the noise edge amplification risk component, and the color shift risk component, respectively. The risk vectors corresponding to multiple image units are arranged according to their spatial positions in the image to be processed, and together they form a multi-parameter coupled risk map.

[0091] After obtaining the risk vector, the adaptive parameter generation module compares the risk vector with a preset risk threshold vector to obtain the risk exceeding the limit vector. Continuing with the above image unit... For example, the preset risk threshold vector is:

[0092]

[0093] The risk exceeding the limit vector is then:

[0094]

[0095] in, This indicates that only the portion exceeding the corresponding threshold is retained, while the risk component not exceeding the threshold is recorded as 0. The adaptive parameter generation module generates a parameter correction coefficient vector based on the risk exceedance vector, and uses the parameter correction coefficient vector to correct the initial processing parameter vector. In one implementation, the parameter correction coefficient vector can be represented as:

[0096]

[0097] After correcting the initial processing parameter vector, the target processing parameter vector is obtained:

[0098]

[0099] in, The five values ​​in the image are used as image units. The target brightness adjustment parameters, target contrast enhancement parameters, target noise reduction parameters, target sharpening parameters, and target color correction parameters.

[0100] In this way, the adaptive parameter generation module does not evaluate the processing results after image enhancement and restoration are completed. Instead, it generates a multi-parameter coupling risk map based on the processing sensitivity vector, the initial processing parameter vector, and the parameter conflict response matrix before performing image processing. Then, it applies reverse constraints to the initial image processing parameter map based on this multi-parameter coupling risk map. This allows for the suppression of risks such as local over-enhancement, noise amplification, texture loss, noise edge amplification, and color shift before the target image data is generated. This makes the target image processing parameter map more suitable for adaptive enhancement and restoration of different image units in complex photographic scenes.

[0101] Example 5

[0102] Based on Examples 1 to 4, this example describes the process by which the image processing output module generates target image data according to the target image processing parameter map.

[0103] The image processing output module obtains the target processing parameter vector corresponding to each image unit, and performs adaptive image enhancement and restoration processing on the corresponding image unit based on the brightness adjustment parameters, contrast enhancement parameters, noise reduction parameters, sharpening parameters, and color correction parameters in the target processing parameter vector. Specifically, the brightness adjustment parameters adjust the local brightness level of the image unit, the contrast enhancement parameters improve the local contrast of the image unit, the noise reduction parameters suppress noise in the image unit, the sharpening parameters enhance the edge and texture details of the image unit, and the color correction parameters correct local color deviations in the image unit.

[0104] When there is an overlapping boundary band between adjacent image units, the image processing output module performs transition fusion on the processing results of adjacent image units within the overlapping boundary band, so that the brightness, contrast, texture intensity and color performance between adjacent image units are continuously transitioned, reducing boundary abrupt changes, local halos or color banding caused by different target processing parameter vectors used by different image units.

[0105] After the above processing, the processing results of each image unit are stitched and fused according to their spatial position in the image to be processed to obtain the target image data. Compared with the image to be processed, the target image data can maintain more stable brightness levels, noise suppression effects, texture details and color naturalness in complex photographic scenes, thereby improving the refinement of image data enhancement and restoration processing.

[0106] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent photographic image data processing system based on AI scene recognition, characterized in that, include: The image acquisition module is used to acquire photographic image data to be processed, normalize the photographic image data to be processed, and divide the normalized photographic image data to be processed into multiple image units to obtain the image to be processed. The scene feature analysis module is used to perform AI scene recognition on the image to be processed, generate a scene region confidence map, and generate a scene sensitivity map based on the scene region confidence map. The scene sensitivity map includes the processing sensitivity vector corresponding to each image unit. The local image quality analysis module is used to extract the brightness state, texture state and noise state of each image unit and generate a local image quality map corresponding to the scene sensitivity map space. The local image quality map includes a brightness distribution map, a texture preservation weight map and a noise estimation map. An adaptive parameter generation module is used to generate an initial image processing parameter map based on the scene sensitivity map and the local image quality map. Based on the processing sensitivity vector, the module weights the parameter conflict relationships among the brightness adjustment parameters, contrast enhancement parameters, noise reduction parameters, sharpening parameters and color correction parameters in the initial image processing parameter map to construct a parameter conflict response matrix corresponding to each image unit. A multi-parameter coupled risk map is generated based on the aforementioned parameter conflict response matrix; The initial image processing parameter map is corrected by reverse constraint based on the multi-parameter coupling risk map to obtain the target image processing parameter map. The image processing output module is used to perform adaptive image enhancement and restoration processing on each image unit in the image to be processed according to the target image processing parameter map, so as to obtain target image data.

2. The intelligent photographic image data processing system based on AI scene recognition according to claim 1, characterized in that: When the image acquisition module performs normalization processing on the photographic image data to be processed, it performs color space unification, pixel value range mapping and image size standardization processing, and divides the normalized photographic image data to be processed into multiple image units according to the preset pixel size. An overlapping boundary band is set between adjacent image units. The overlapping boundary band is used for the image processing output module to perform transition fusion on the processing results of adjacent image units.

3. The intelligent photographic image data processing system based on AI scene recognition according to claim 1, characterized in that: The scene region confidence map includes a scene confidence vector corresponding to each image unit. The scene confidence vector includes the confidence values ​​of the corresponding image unit belonging to the highlight region, dark region, subject region, background region, and high texture region, respectively.

4. The intelligent photographic image data processing system based on AI scene recognition according to claim 1, characterized in that: The scene feature analysis module generates a processing sensitivity vector for each image unit based on the scene confidence vector and a preset scene processing sensitivity mapping relationship. The processing sensitivity vector includes brightness adjustment sensitivity, contrast enhancement sensitivity, noise reduction sensitivity, sharpening sensitivity, and color correction sensitivity.

5. The intelligent photographic image data processing system based on AI scene recognition according to claim 1, characterized in that: When generating the local image quality map, the local image quality analysis module calculates the average brightness value, the proportion of bright pixels and the proportion of dark pixels of each image unit to generate the brightness distribution map, calculates the gradient intensity and edge density of each image unit to generate the texture preservation weight map, and generates the noise estimation map based on the pixel fluctuation amplitude of the flat sub-region within each image unit.

6. The intelligent photographic image data processing system based on AI scene recognition according to claim 1, characterized in that: For each image unit, the adaptive parameter generation module combines the processing sensitivity vector, brightness distribution value, texture preservation weight value, and noise estimation value corresponding to that image unit to generate an initial processing parameter vector for that image unit. The initial processing parameter vector includes brightness adjustment parameters, contrast enhancement parameters, noise reduction parameters, sharpening parameters, and color correction parameters. The initial processing parameter vectors corresponding to each image unit together constitute the initial image processing parameter map.

7. The intelligent photographic image data processing system based on AI scene recognition according to claim 6, characterized in that: The parameter conflict response matrix is ​​a matrix that corresponds one-to-one with each image unit. The rows and columns of the parameter conflict response matrix correspond to the brightness adjustment parameter, contrast enhancement parameter, noise reduction parameter, sharpening parameter, and color correction parameter, respectively. The matrix elements in the parameter conflict response matrix are used to characterize the influence intensity of the processing risk caused by different image processing parameters.

8. The intelligent photographic image data processing system based on AI scene recognition according to claim 7, characterized in that: When constructing the parameter conflict response matrix, the adaptive parameter generation module uses the processing sensitivity vector corresponding to each image unit as a weighting factor to perform weighting processing on the preset parameter conflict basis matrix to obtain the parameter conflict response matrix of the corresponding image unit. The parameter conflict basis matrix includes the conflict relationship between brightness adjustment parameters and noise amplification risk, the conflict relationship between contrast enhancement parameters and local over-enhancement risk, the conflict relationship between noise reduction parameters and texture loss risk, the conflict relationship between sharpening parameters and noise edge amplification risk, and the conflict relationship between color correction parameters and color shift risk.

9. The intelligent photographic image data processing system based on AI scene recognition according to claim 8, characterized in that: The adaptive parameter generation module generates a risk vector for each image unit based on the initial processing parameter vector and parameter conflict response matrix corresponding to each image unit. The risk vector includes a local over-enhancement risk component, a noise amplification risk component, a texture loss risk component, a noise edge amplification risk component, and a color shift risk component. The risk vectors corresponding to each image unit together constitute the multi-parameter coupled risk map.

10. The intelligent photographic image data processing system based on AI scene recognition according to claim 9, characterized in that: When performing reverse constraint correction, the adaptive parameter generation module compares the risk vector of each image unit with the preset risk threshold vector to obtain the risk exceeding the limit vector, and generates a parameter correction coefficient vector based on the risk exceeding the limit vector; it then corrects the initial processing parameter vector of the image unit based on the parameter correction coefficient vector to obtain the target processing parameter vector of the image unit. The target processing parameter vectors of each image unit together constitute the target image processing parameter map.

Images