Image data processing method and system for an image processing chip

Abstract

The present invention relates to an image data processing method and system for an image processing chip, and relates to the technical field of image processing. The method includes: obtaining a plurality of sample image groups, including a sample image to be enhanced and an enhanced sample image corresponding to the sample image to be enhanced; for each sample image group, extracting a first image feature corresponding to the sample image to be enhanced in the sample image group, and extracting a second image feature corresponding to the enhanced sample image in the sample image group; establishing a plurality of image enhancement models based on the first image feature and the second image feature of each sample image group; extracting a third image feature corresponding to the image to be enhanced; determining a target image enhancement model from the plurality of image enhancement models based on the third image feature corresponding to the image to be enhanced and the first image feature of each sample image group; and generating an enhanced image corresponding to the image to be enhanced through the target image enhancement model, which has the advantages of improving the efficiency and effect of image enhancement.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image data processing method and system of an image processing chip. Background Art

[0002] An image processing chip is an integrated circuit chip specially used for processing image data. It has the advantages of fast processing speed, low energy consumption and high performance. Image processing chips are widely used in computer vision, artificial intelligence, medical imaging and other fields. Image processing chips have high-performance processing capabilities and can realize high-speed processing and analysis of images. By utilizing the ability of parallel computing, it can process multiple pixels at the same time and can realize the calculation of multiple complex algorithms. For example, in the field of computer vision, image processing chips can realize tasks such as image recognition, target detection, and image segmentation. Image processing chips can achieve low energy consumption and high efficiency processing. Due to its dedicated design and fine optimization, image processing chips can minimize energy consumption and improve energy utilization efficiency when processing image data. Image processing chips also have high stability and reliability. It has been rigorously tested and verified during the design and manufacturing process, and can run stably in various working environments and application scenarios without prone to failure.

[0003] Due to the complexity of conditions, especially at night, video images are often low in brightness, high in noise, and low in contrast, making it difficult to achieve ideal acquisition effects and obtain the desired information. Therefore, it is necessary to enhance the images formed under these low-light conditions. The purpose of low-light image enhancement is to highlight key information, eliminate interference information as much as possible, and make the original unclear or low-brightness images clear or enhance the brightness. Under low-light conditions, due to the uncertainty of lighting conditions, the captured images are of poor quality and are mixed with noise, resulting in a serious decrease in contrast and brightness, which affects the details of the dark areas of the image and causes color deviation.

[0004] Therefore, there is a need for an image data processing method and system for an image processing chip to improve the efficiency and effect of image enhancement. Summary of the invention

[0005] The present invention provides an image data processing method for an image processing chip, which is applied to the image processing chip and includes: obtaining a plurality of sample image groups, where each sample image group includes a sample image to be enhanced and an enhanced sample image corresponding to the sample image to be enhanced, and the light intensity corresponding to the sample image to be enhanced is lower than the light intensity corresponding to the enhanced sample image; for each sample image group, extracting a first image feature corresponding to the sample image to be enhanced in the sample image group, and extracting a second image feature corresponding to the enhanced sample image in the sample image group; based on the first image feature and the second image feature of each sample image group, establishing a plurality of image enhancement models; obtaining an image to be enhanced; extracting a third image feature corresponding to the image to be enhanced; based on the third image feature corresponding to the image to be enhanced and the first image feature of each sample image group, determining a target image enhancement model from the plurality of image enhancement models; and generating an enhanced image corresponding to the image to be enhanced through the target image enhancement model.

[0006] Further, extracting the first image feature corresponding to the sample image to be enhanced in the sample image group includes: determining a plurality of candidate image feature factors; based on the plurality of candidate image feature factors, extracting a first test image feature corresponding to the sample image to be enhanced in each sample image group; for any two of the sample images to be enhanced, determining the test image feature similarity of the two sample images to be enhanced based on the first test image features corresponding to the two sample images to be enhanced; based on the test image feature similarity of the two sample images to be enhanced, screening the plurality of candidate image feature factors to determine a plurality of first target image feature factors; and based on the plurality of first target image feature factors, extracting the first image feature corresponding to the sample image to be enhanced in the sample image group.

[0007] Further, screening the plurality of candidate image feature factors based on the test image feature similarity of the two sample images to be enhanced to determine a plurality of first target image feature factors includes: for each candidate image feature factor, calculating a first screening parameter corresponding to the candidate image feature factor based on the test image feature similarity of the two sample images to be enhanced; for any two of the candidate image feature factors, calculating a first correlation parameter of the two candidate image feature factors based on the first test image feature corresponding to the sample image to be enhanced in each sample image group; and screening the plurality of candidate image feature factors according to the first screening parameter corresponding to each candidate image feature factor and the first correlation parameter of any two of the candidate image feature factors to determine a plurality of first target image feature factors.

[0008] Further, extracting the second image features corresponding to the enhanced sample images of the sample image group includes: based on the multiple candidate image feature factors, extracting the second test image features corresponding to the enhanced sample images of each sample image group; for any two of the enhanced sample images, determining the test image feature similarity of the two enhanced sample images based on the second test image features corresponding to the two enhanced sample images; for each candidate image feature factor, calculating a second screening parameter corresponding to the candidate image feature factor based on the test image feature similarity of the two enhanced sample images; for any two of the candidate image feature factors, calculating a second correlation parameter of the two candidate image feature factors based on the second test image features corresponding to the enhanced sample images of each sample image group; screening the multiple candidate image feature factors according to the second screening parameter corresponding to each candidate image feature factor and the second correlation parameter of any two candidate image feature factors to determine multiple second target image feature factors; and extracting the second image features corresponding to the enhanced sample images of the sample image group based on the multiple second target image feature factors.

[0009] Further, establishing multiple image enhancement models based on the first image features and the second image features of each sample image group includes: for any two of the sample image groups, calculating the image group feature similarity of the two sample image groups based on the first image features and the second image features of each sample image group; clustering the multiple sample image groups according to the image group feature similarity of any two sample image groups to determine multiple enhanced image classes; and establishing the multiple image enhancement models according to the multiple enhanced image classes.

[0010] Further, extracting the third image features corresponding to the image to be enhanced includes: extracting the third image features corresponding to the image to be enhanced based on the multiple first target image feature factors.

[0011] Further, determining a target image enhancement model from the multiple image enhancement models based on the third image features corresponding to the image to be enhanced and the first image features of each sample image group includes: calculating the class matching degree of the image to be enhanced and each enhanced image class based on the third image features corresponding to the image to be enhanced and the first image features of each sample image group; determining a target enhanced image class according to the class matching degree of the image to be enhanced and each enhanced image class; and determining a target image enhancement model from the multiple image enhancement models according to the target enhanced image class.

[0012] Further, a target enhanced image class is determined according to the class matching degree between the image to be enhanced and each of the enhanced image classes, including: determining the target enhanced image class according to the class matching degree between the image to be enhanced and each of the enhanced image classes and the class similarity between any two of the enhanced image classes.

[0013] Further, the target image enhancement model includes an image segmentation unit, a brightness adjustment unit, a contrast adjustment unit, a noise removal unit, an artifact removal unit, a color saturation adjustment unit, an edge enhancement unit, a texture enhancement unit, and an image synthesis unit. Among them, the image segmentation unit is used to segment the image to be enhanced to generate a plurality of regional images; the noise removal unit is used to denoise the plurality of regional images to generate a plurality of denoised regional images; the brightness adjustment unit is used to adjust the brightness of the plurality of denoised regional images to generate a plurality of regional images with adjusted brightness; the color saturation adjustment unit is used to adjust the color saturation of the plurality of regional images with adjusted brightness to generate a plurality of regional images with adjusted color saturation; the artifact removal unit is used to remove the artifacts of the plurality of regional images with adjusted color saturation to generate a plurality of regional images with removed artifacts; the image synthesis unit is used to generate an intermediate image based on the plurality of regional images with removed artifacts; the edge enhancement unit is used to enhance the edge information of the intermediate image to generate an intermediate image with enhanced edges; the texture enhancement unit is used to enhance the texture of the intermediate image with enhanced edges to generate an enhanced image corresponding to the image to be enhanced.

[0014] The present invention provides an image data processing system for an image processing chip, which is used to execute the above-mentioned image data processing method for an image processing chip and runs on the image processing chip. The system includes: a sample acquisition module, configured to acquire a plurality of sample image groups, where each sample image group includes a sample image to be enhanced and an enhanced sample image corresponding to the sample image to be enhanced, and the light intensity corresponding to the sample image to be enhanced is lower than the light intensity corresponding to the enhanced sample image; a feature extraction module, configured to, for each sample image group, extract a first image feature corresponding to the sample image to be enhanced in the sample image group, and extract a second image feature corresponding to the enhanced sample image in the sample image group; a model establishment module, configured to establish a plurality of image enhancement models based on the first image feature and the second image feature of each sample image group; an image acquisition module, configured to acquire an image to be enhanced; the feature extraction module is further configured to extract a third image feature corresponding to the image to be enhanced; a model determination module, configured to determine a target image enhancement model from the plurality of image enhancement models based on the third image feature corresponding to the image to be enhanced and the first image feature of each sample image group; and an image processing module, configured to generate an enhanced image corresponding to the image to be enhanced through the target image enhancement model.

[0015] Compared with the prior art, the image data processing method and system for an image processing chip provided in this specification have at least the following beneficial effects:

[0016] 1. By establishing an image enhancement model, this method can enhance images with low light intensity, improve their lighting conditions, and thus enhance the clarity and visibility of the images. This is particularly important for images taken in low-light environments and can significantly enhance the details and color performance of the images. Based on machine learning or deep learning techniques, multiple image enhancement models are trained to automatically identify and enhance images. This reduces the need for manual intervention and improves the efficiency and accuracy of image processing. Multiple sample image groups are used for training, so it can learn image features under different lighting conditions. This makes the method more adaptable and robust when processing images with different lighting conditions;

[0017] 2. By calculating the screening parameters and correlation parameters of candidate image feature factors, various candidate image feature factors are screened to determine multiple target image feature factors. This helps to optimize the image feature extraction process and improve the effectiveness and accuracy of image features. At the same time, by clustering the sample image groups and determining multiple enhanced image classes, multiple image enhancement models can be established according to different image features. This increases the diversity and adaptability of the models, enabling the models to better handle different types of image enhancement tasks. Description of the Drawings

[0018] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same numbers represent the same structures, where:

[0019] Figure 1 is a flowchart of an image data processing method of an image processing chip shown in an embodiment of the present application;

[0020] Figure 2 is a schematic structural diagram of a target image enhancement model shown in an embodiment of the present application;

[0021] Figure 3 is a module diagram of an image data processing system of an image processing chip shown in an embodiment of the present application. Detailed implementation manners

[0022] To more clearly illustrate the technical solutions of the embodiments of this specification, the accompanying drawings required for the description of the embodiments will be briefly introduced below.

[0023] An image data processing method of an image processing chip can be applied to an image processing chip.

[0024] Preferably, the image processing chip can be a NIO Core ISP image processing chip, which supports multi-format intelligent image conversion, receives high-resolution and high-frame-rate display images sent from a main control such as a mobile phone, a game console, a tablet or an industrial CPU, performs scaling, rotation or frame rate conversion, and outputs images that meet the specification requirements. In addition to display conversion, a high-performance MCU is built-in to meet various control and protocol conversions, such as dimming, touch conversion, external control, etc. At the same time, a high-security encryption module, real-time data monitoring, a large-capacity XRAM, direct display output, a large-capacity SPI Flash, an ADC analog module, etc. are built-in, which are applicable to scenarios such as display terminals, image conversion, interface conversion, and edge acceleration.

[0025] Figure 1 is a flowchart of an image data processing method of an image processing chip shown in an embodiment of the present application, as Figure 1 shown, an image data processing method of an image processing chip may include the following steps.

[0026] Step 110, obtain a plurality of sample image groups, where each sample image group includes a sample image to be enhanced and an enhanced sample image corresponding to the sample image to be enhanced, and the illumination intensity corresponding to the sample image to be enhanced is lower than the illumination intensity corresponding to the enhanced sample image.

[0027] Specifically, the sample image to be enhanced can be an image acquired under low light intensity, and the enhanced sample image can be an image acquired under high light intensity.

[0028] Step 120, for each sample image group, extract the first image feature corresponding to the sample image to be enhanced in the sample image group, and extract the second image feature corresponding to the enhanced sample image in the sample image group.

[0029] Preferably, extracting the first image feature corresponding to the sample image to be enhanced in the sample image group includes:

[0030] Determine multiple candidate image feature factors, for example, brightness feature, contrast feature, detail feature, color feature, etc.;

[0031] Based on the multiple candidate image feature factors, extract the first test image feature corresponding to the sample image to be enhanced in each sample image group, where the first test image feature includes the value corresponding to each candidate image feature factor extracted from the sample image to be enhanced;

[0032] For any two sample images to be enhanced, based on the first test image features corresponding to the two sample images to be enhanced, determine the test image feature similarity between the two sample images to be enhanced;

[0033] Based on the test image feature similarity between the two sample images to be enhanced, screen the multiple candidate image feature factors to determine multiple first target image feature factors;

[0034] Based on the multiple first target image feature factors, extract the first image feature corresponding to the sample image to be enhanced in the sample image group, where the first image feature includes the value corresponding to each first target image feature factor extracted from the sample image to be enhanced.

[0035] Specifically, the test image feature similarity between two sample images to be enhanced can be calculated according to the following formula:

[0036] ;

[0037] where, is the test image feature similarity between the i-th sample image to be enhanced and the j-th sample image to be enhanced, is a preset parameter, greater than 0, is the value of the i-th sample image to be enhanced corresponding to the k-th candidate image feature factor, is the value of the j-th sample image to be enhanced corresponding to the k-th candidate image feature factor, is the total number of candidate image feature factors.

[0038] Preferably, based on the similarity of test image features of two sample images to be enhanced, multiple candidate image feature factors are screened to determine multiple first target image feature factors, including:

[0039] For each candidate image feature factor, based on the similarity of test image features of two sample images to be enhanced, calculate the first screening parameter corresponding to the candidate image feature factor;

[0040] For any two candidate image feature factors, based on the first test image features corresponding to the sample images to be enhanced in each sample image group, calculate the first correlation parameter between the two candidate image feature factors;

[0041] According to the first screening parameter corresponding to each candidate image feature factor and the first correlation parameter between any two candidate image feature factors, screen multiple candidate image feature factors to determine multiple first target image feature factors.

[0042] Specifically, the first screening parameter corresponding to the candidate image feature factor can be calculated according to the following formula:

[0043] ;

[0044] where is the first screening parameter corresponding to the k-th candidate image feature factor, is the total number of sample images to be enhanced, is the similarity of test image features between the e-th sample image to be enhanced and the j-th sample image to be enhanced.

[0045] The first correlation parameter between two candidate image feature factors can be calculated according to the following formula:

[0046] ;

[0047] where is the first correlation parameter between the k-th candidate image feature factor and the t-th candidate image feature factor, is the value of the t-th candidate image feature factor corresponding to the i-th sample image to be enhanced.

[0048] The candidate image feature factor with the first screening parameter greater than the first screening parameter threshold can be used as the first factor of interest, and the candidate image feature factor with the first correlation parameter greater than the first correlation parameter threshold with the first factor of interest and not used as the first factor of interest can be used as the second factor of interest. The multiple first target image feature factors can include each first factor of interest and each second factor of interest.

[0049] Preferably, extract the second image features corresponding to the enhanced sample images of the sample image group, including:

[0050] Based on multiple candidate image feature factors, extract the second test image features corresponding to the enhanced sample images of each sample image group, where the second test image features include the values corresponding to each candidate image feature factor extracted from the enhanced sample images;

[0051] For any two enhanced sample images, based on the second test image features corresponding to the two enhanced sample images, determine the test image feature similarity of the two enhanced sample images. The method for calculating the test image feature similarity of the two enhanced sample images is similar to the method for calculating the test image feature similarity of the two sample images to be enhanced, which will not be elaborated here;

[0052] For each candidate image feature factor, based on the test image feature similarity of the two enhanced sample images, calculate the second screening parameter corresponding to the candidate image feature factor. The method for calculating the second screening parameter corresponding to the candidate image feature factor is similar to the method for calculating the first screening parameter corresponding to the candidate image feature factor, which will not be elaborated here;

[0053] For any two candidate image feature factors, based on the second test image features corresponding to the enhanced sample images of each sample image group, calculate the second correlation parameter of the two candidate image feature factors. The method for calculating the second correlation parameter of the two candidate image feature factors is similar to the method for calculating the first correlation parameter of the two candidate image feature factors, which will not be elaborated here;

[0054] According to the second screening parameter corresponding to each candidate image feature factor and the second correlation parameter of any two candidate image feature factors, screen the multiple candidate image feature factors to determine multiple second target image feature factors. The method for determining multiple second target image feature factors is similar to the method for determining multiple first target image feature factors, which will not be elaborated here;

[0055] Based on the multiple second target image feature factors, extract the second image features corresponding to the enhanced sample images of the sample image group, where the second image features include the values corresponding to each second target image feature factor extracted from the enhanced sample images.

[0056] Step 130, based on the first image features and the second image features of each sample image group, establish multiple image enhancement models.

[0057] Specifically include:

[0058] For any two sample image groups, based on the first image features and the second image features of each sample image group, calculate the image group feature similarity of the two sample image groups;

[0059] Cluster multiple sample image groups according to the image group feature similarity between any two sample image groups, and determine multiple enhanced image classes;

[0060] Establish multiple image enhancement models according to the multiple enhanced image classes, where the image enhancement model can be a Convolutional Neural Network (CNN) model.

[0061] Specifically, for any two sample image groups, the first image feature similarity can be calculated according to the first image feature of each sample image group, and the second image feature similarity can be calculated according to the second image feature of each sample image group. The first image feature similarity and the second image feature similarity can be weighted to obtain the image group feature similarity of the two sample image groups. The methods of calculating the first image feature similarity and the second image feature similarity are similar to those of calculating the test image feature similarity of two sample images to be enhanced, which will not be elaborated here.

[0062] The multiple sample image groups can be clustered by a clustering algorithm (such as the K-means clustering algorithm, hierarchical clustering algorithm, etc.) to determine multiple enhanced image classes, and one enhanced image class corresponds to one clustering cluster. One enhanced image class corresponds to one clustering cluster.

[0063] Figure 2 It is a schematic structural diagram of the target image enhancement model shown in an embodiment of the present application, as Figure 2 shown. Preferably, the target image enhancement model includes an image segmentation unit, a brightness adjustment unit, a contrast adjustment unit, a noise removal unit, an artifact removal unit, a color saturation adjustment unit, an edge enhancement unit, a texture enhancement unit, and an image synthesis unit. Among them, the image segmentation unit is used to segment the image to be enhanced to generate multiple regional images, the noise removal unit is used to denoise the multiple regional images to generate denoised multiple regional images, the brightness adjustment unit is used to adjust the brightness of the denoised multiple regional images to generate brightness-adjusted multiple regional images, the color saturation adjustment unit is used to adjust the color saturation of the brightness-adjusted multiple regional images to generate color saturation-adjusted multiple regional images, the artifact removal unit is used to remove the artifacts of the color saturation-adjusted multiple regional images to generate artifact-removed multiple regional images, the image synthesis unit is used to generate an intermediate image based on the artifact-removed multiple regional images, the edge enhancement unit is used to enhance the edge information of the intermediate image to generate an edge-enhanced intermediate image, and the texture enhancement unit is used to enhance the texture of the edge-enhanced intermediate image to generate the enhanced image corresponding to the image to be enhanced.

[0064] Step 140, obtain the image to be enhanced.

[0065] Specifically, the image to be enhanced is an image collected under low light intensity.

[0066] Step 150, extract the third image feature corresponding to the image to be enhanced.

[0067] Specifically, it includes:

[0068] Based on multiple first target image feature factors, extract the third image feature corresponding to the image to be enhanced, where the third image feature corresponding to the image to be enhanced may include the values corresponding to each first target image feature factor extracted from the image to be enhanced.

[0069] Step 160, determine the target image enhancement model from multiple image enhancement models based on the third image feature corresponding to the image to be enhanced and the first image feature of each sample image group.

[0070] Specifically, it includes:

[0071] Based on the third image feature corresponding to the image to be enhanced and the first image feature of each sample image group, calculate the class matching degree between the image to be enhanced and each enhanced image class;

[0072] According to the class matching degree between the image to be enhanced and each enhanced image class, determine the target enhanced image class;

[0073] According to the target enhanced image class, determine the target image enhancement model from multiple image enhancement models.

[0074] Preferably, according to the class matching degree between the image to be enhanced and each enhanced image class, determining the target enhanced image class includes:

[0075] According to the class matching degree between the image to be enhanced and each enhanced image class and the class similarity between any two enhanced image classes, determine the target enhanced image class.

[0076] Specifically, for each enhanced image class, multiple images to be enhanced can be sampled from the enhanced image class. According to the first image feature of the sampled images to be enhanced and the third image feature corresponding to the image to be enhanced, calculate the feature similarity between the image to be enhanced and the sampled images to be enhanced. The method is similar to calculating the test image feature similarity between two sample images to be enhanced, which will not be elaborated here. Calculate the average value of the feature similarities between the image to be enhanced and each sampled image to be enhanced as the class matching degree between the image to be enhanced and the enhanced image class.

[0077] The class matching degree between the image to be enhanced and each enhanced image class can be calculated according to the following formula:

[0078] ;

[0079] Where, is the class matching degree between the image to be enhanced and the g-th enhanced image class, is the total number of images to be enhanced sampled from the g-th enhanced image class, is the feature similarity between the image to be enhanced and the m-th image to be enhanced sampled from the g-th enhanced image class.

[0080] According to the class matching degree between the image to be enhanced and each enhanced image class and the class similarity between any two enhanced image classes, calculate the class screening value of each enhanced image class, and use the enhanced image class with the largest class screening value as the target enhanced image class.

[0081] For any two enhanced image classes, multiple images to be enhanced can be sampled from each of the two enhanced image classes, calculate the feature similarity between each image to be enhanced sampled from one enhanced image class and each image to be enhanced sampled from the other enhanced image class. Similar to the method of calculating the feature similarity of the test image of two images to be enhanced samples, it will not be elaborated here. Take the average of the feature similarities between each image to be enhanced sampled from one enhanced image class and each image to be enhanced sampled from the other enhanced image class as the class similarity between the two enhanced image classes.

[0082] The class screening value of the enhanced image class can be calculated according to the following formula:

[0083] ;

[0084] where, is the class screening value of the g-th enhanced image class, is a preset parameter, greater than 0, is the class matching degree between the enhanced image and the h-th enhanced image class, is the class similarity between the g-th enhanced image class and the h-th enhanced image class, is the total number of enhanced image classes.

[0085] Step 170, generate the enhanced image corresponding to the image to be enhanced through the target image enhancement model.

[0086] Figure 3 is a module diagram of an image data processing system of an image processing chip shown in an embodiment of the present application, as Figure 3 shown, an image data processing system of an image processing chip may include a sample acquisition module, a feature extraction module, a model establishment module, an image acquisition module, a model determination module, and an image processing module.

[0087] The sample acquisition module can be used to acquire multiple groups of sample images, where each group of sample images includes a sample image to be enhanced and an enhanced sample image corresponding to the sample image to be enhanced, and the illumination intensity corresponding to the sample image to be enhanced is lower than the illumination intensity corresponding to the enhanced sample image;

[0088] The feature extraction module can be used to, for each group of sample images, extract the first image feature corresponding to the sample image to be enhanced in the group of sample images, and extract the second image feature corresponding to the enhanced sample image in the group of sample images;

[0089] The model establishment module can be used to establish multiple image enhancement models based on the first image feature and the second image feature of each group of sample images;

[0090] The image acquisition module can be used to acquire an image to be enhanced;

[0091] The feature extraction module can also be used to extract the third image feature corresponding to the image to be enhanced;

[0092] The model determination module can be used to determine a target image enhancement model from multiple image enhancement models based on the third image feature corresponding to the image to be enhanced and the first image feature of each group of sample images;

[0093] The image processing module can be used to generate an enhanced image corresponding to the image to be enhanced through the target image enhancement model.

[0094] An image data processing system of an image processing chip can be used to execute an image data processing method of an image processing chip, running on the image processing chip. For more descriptions of the image data processing system of an image processing chip, reference can be made to the relevant descriptions of the image data processing method of an image processing chip, which will not be elaborated here.

[0095] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other deformations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification can be regarded as consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. An image data processing method for an image processing chip, applied to an image processing chip, characterized in that: include: Acquire multiple sample image groups, wherein the sample image groups include sample images to be enhanced and enhanced sample images corresponding to the sample images to be enhanced, and the light intensity corresponding to the sample images to be enhanced is lower than the light intensity corresponding to the enhanced sample images; For each sample image group, extracting a first image feature corresponding to a sample image to be enhanced of the sample image group, and extracting a second image feature corresponding to an enhanced sample image of the sample image group, including: determining a plurality of candidate image feature factors, taking a candidate image feature factor whose first screening parameter is greater than a first screening parameter threshold as a first factor of interest, taking a candidate image feature factor whose first association parameter with the first factor of interest is greater than the first association parameter threshold and is not used as a first factor of interest as a second factor of interest, the plurality of first target image feature factors including each first factor of interest and each second factor of interest, and extracting a first image feature corresponding to the sample image to be enhanced of the sample image group based on the plurality of first target image feature factors; Based on the first image feature and the second image feature of each sample image group, multiple image enhancement models are established, including: for any two sample image groups, according to the first image feature of each sample image group, the first image feature similarity is calculated, according to the second image feature of each sample image group, the second image feature similarity is calculated, the first image feature similarity and the second image feature similarity are weighted to obtain the image group feature similarity of the two sample image groups, according to the image group feature similarity of any two sample image groups, the multiple sample image groups are clustered to determine multiple enhanced image classes, and the multiple image enhancement models are established according to the multiple enhanced image classes; Acquire an image to be enhanced; Extracting a third image feature corresponding to the image to be enhanced; Based on the third image feature corresponding to the image to be enhanced and the first image feature of each sample image group, determining a target image enhancement model from multiple image enhancement models, including: calculating a class screening value of each enhanced image class according to a class matching degree between the image to be enhanced and each enhanced image class and a class similarity between any two enhanced image classes, taking the enhanced image class with the largest class screening value as the target enhanced image class, and determining the target image enhancement model from multiple image enhancement models according to the target enhanced image class; An enhanced image corresponding to the image to be enhanced is generated by the target image enhancement model.

2. The image data processing method of an image processing chip according to claim 1, characterized in that: Extracting a first image feature corresponding to the sample image to be enhanced of the sample image group includes: Determine a plurality of candidate image feature factors; Extracting a first test image feature corresponding to the sample image to be enhanced of each of the sample image groups based on the multiple candidate image feature factors; For any two of the sample images to be enhanced, based on the first test image features corresponding to the two sample images to be enhanced, determining the similarity of the test image features of the two sample images to be enhanced; Based on the similarity of the test image features of the two sample images to be enhanced, the plurality of candidate image feature factors are screened to determine a plurality of first target image feature factors; Based on the multiple first target image feature factors, first image features corresponding to the sample images to be enhanced in the sample image group are extracted.

3. The image data processing method of an image processing chip according to claim 2, characterized in that: Based on the similarity of the test image features of the two sample images to be enhanced, the plurality of candidate image feature factors are screened to determine a plurality of first target image feature factors, including: For each of the candidate image feature factors, based on the similarity of the test image features of the two sample images to be enhanced, calculating a first screening parameter corresponding to the candidate image feature factor; For any two of the candidate image feature factors, based on the first test image feature corresponding to the sample image to be enhanced of each of the sample image groups, calculate the first correlation parameter of the two candidate image feature factors; The plurality of candidate image feature factors are screened according to the first screening parameter corresponding to each of the candidate image feature factors and the first associated parameters of any two of the candidate image feature factors to determine a plurality of first target image feature factors.

4. The image data processing method of an image processing chip according to claim 2, characterized in that: Extracting a second image feature corresponding to the enhanced sample image of the sample image group includes: Extracting second test image features corresponding to the enhanced sample images of each sample image group based on the multiple candidate image feature factors; For any two of the enhanced sample images, based on the second test image features corresponding to the two enhanced sample images, determine the similarity of the test image features of the two enhanced sample images; For each of the candidate image feature factors, based on the test image feature similarity of the two enhanced sample images, calculate a second screening parameter corresponding to the candidate image feature factor; For any two of the candidate image feature factors, based on the second test image features corresponding to the enhanced sample images of each of the sample image groups, calculate second correlation parameters of the two candidate image feature factors; Screening the plurality of candidate image feature factors according to the second screening parameter corresponding to each of the candidate image feature factors and the second associated parameters of any two of the candidate image feature factors to determine a plurality of second target image feature factors; Based on the multiple second target image feature factors, second image features corresponding to the enhanced sample images of the sample image group are extracted.

5. The image data processing method of an image processing chip according to any one of claims 2 to 4, characterized in that: Based on the first image feature and the second image feature of each of the sample image groups, multiple image enhancement models are established, including: For any two of the sample image groups, based on the first image feature and the second image feature of each of the sample image groups, calculating the image group feature similarity of the two sample image groups; Clustering the plurality of sample image groups according to the image group feature similarity of any two of the sample image groups to determine a plurality of enhanced image classes; The multiple image enhancement models are established according to the multiple enhanced image classes.

6. The image data processing method of an image processing chip according to any one of claims 2 to 4, characterized in that: Extracting a third image feature corresponding to the image to be enhanced includes: Based on the multiple first target image feature factors, a third image feature corresponding to the image to be enhanced is extracted.

7. The image data processing method of an image processing chip according to claim 5, characterized in that: Determining a target image enhancement model from the plurality of image enhancement models based on a third image feature corresponding to the image to be enhanced and a first image feature of each of the sample image groups comprises: Calculating a class matching degree between the image to be enhanced and each enhanced image class based on a third image feature corresponding to the image to be enhanced and a first image feature of each of the sample image groups; Determining a target enhanced image class according to a class matching degree between the image to be enhanced and each enhanced image class; According to the target enhanced image class, a target image enhancement model is determined from the multiple image enhancement models.

8. The image data processing method of an image processing chip according to claim 7, characterized in that: Determining a target enhanced image class according to a class matching degree between the image to be enhanced and each enhanced image class includes: The target enhanced image class is determined according to the class matching degree between the image to be enhanced and each enhanced image class and the class similarity between any two enhanced image classes.

9. The image data processing method of an image processing chip according to any one of claims 1 to 4, characterized in that: The target image enhancement model includes an image segmentation unit, a brightness adjustment unit, a contrast adjustment unit, a noise removal unit, an artifact removal unit, a color saturation adjustment unit, an edge enhancement unit, a texture enhancement unit and an image synthesis unit, wherein the image segmentation unit is used to segment the image to be enhanced to generate a plurality of region images, the noise removal unit is used to denoise the plurality of region images to generate a plurality of denoised region images, the brightness adjustment unit is used to adjust the brightness of the plurality of denoised region images to generate a plurality of region images after brightness adjustment, the color saturation adjustment unit is used to adjust the color saturation of the plurality of region images after brightness adjustment to generate a plurality of region images after color saturation adjustment, the artifact removal unit is used to remove artifacts from the plurality of region images after color saturation adjustment to generate a plurality of region images after artifact removal, the image synthesis unit is used to generate an intermediate image based on the plurality of region images after artifact removal, the edge enhancement unit is used to enhance the edge information of the intermediate image to generate an edge-enhanced intermediate image, and the texture enhancement unit is used to enhance the texture of the edge-enhanced intermediate image to generate an enhanced image corresponding to the image to be enhanced.

10. An image data processing system of an image processing chip, characterized in that: The method for processing image data of an image processing chip according to any one of claims 1 to 9 is used to execute the method, which runs on the image processing chip, and comprises: A sample acquisition module, used to acquire a plurality of sample image groups, wherein the sample image groups include a sample image to be enhanced and an enhanced sample image corresponding to the sample image to be enhanced, and the light intensity corresponding to the sample image to be enhanced is lower than the light intensity corresponding to the enhanced sample image; A feature extraction module, used for extracting, for each of the sample image groups, first image features corresponding to the sample images to be enhanced in the sample image group, and second image features corresponding to the enhanced sample images in the sample image group; A model building module, used for building a plurality of image enhancement models based on the first image feature and the second image feature of each of the sample image groups; An image acquisition module, used for acquiring an image to be enhanced; The feature extraction module is also used to extract a third image feature corresponding to the image to be enhanced; A model determination module, configured to determine a target image enhancement model from the plurality of image enhancement models based on a third image feature corresponding to the image to be enhanced and a first image feature of each of the sample image groups; The image processing module is used to generate an enhanced image corresponding to the image to be enhanced through the target image enhancement model.

Images