Method and apparatus for tuning an image signal processor

Abstract

Provided are a method and device for tuning an image signal processor. The method for tuning the image signal processor comprises: obtaining an adjusted image of an image by adjusting the size of the image, and obtaining a plurality of image blocks of the image by performing a tiling process on the image, wherein the image is an output image of the image signal processor obtained in response to an input image being input into the image signal processor; extracting a brightness feature and a contrast feature of the image based on the adjusted image; extracting a noise feature and a texture feature of the image based on the plurality of image blocks; obtaining a feature extraction result of the image by fusing the brightness feature, the contrast feature, the noise feature and the texture feature; and tuning the image signal processor based on the feature extraction result and a current parameter of the image signal processor, so as to improve the tuning effect of the image signal processor by improving the accuracy and effectiveness of feature extraction.

Description

Technical Field

[0001] This disclosure relates to the field of image signal processing technology. More specifically, this disclosure relates to a method and apparatus for tuning the parameters of an image signal processor, a computer-readable storage medium, and a computing device. Background Technology

[0002] While deep learning algorithms can achieve hyperparameter tuning, their performance suffers significantly when faced with the hyperparameter tuning challenges of commercial ISPs due to the large number of parameters and modules. Furthermore, the extensive time required for network training makes them unsuitable for commercial ISPs. Additionally, many hyperparameter tuning methods employ pre-trained neural networks as feature extractors. Since these networks are trained for classification tasks, they primarily extract semantically relevant features. These semantically relevant features are relatively redundant for ISP hyperparameter tuning, as low-level visual tasks require lower-level visual features. In conclusion, the poor effectiveness and efficiency of feature extraction in these techniques lead to unsatisfactory hyperparameter tuning results. Summary of the Invention

[0003] Exemplary embodiments of this disclosure provide a method and apparatus, a computer-readable storage medium, and a computing device for tuning the parameters of an image signal processor, in order to solve the problem in the related art where the parameter tuning effect of the image signal processor is poor due to the poor effect and effectiveness of feature extraction.

[0004] According to exemplary embodiments of the present disclosure, a method for parameter tuning of an image signal processor is provided, comprising: obtaining an adjusted image of the image by adjusting the size of the image, and obtaining a plurality of image blocks of the image by slicing the image into blocks, wherein the image is an output image of the image signal processor obtained in response to an input image being input into the image signal processor; extracting brightness features and contrast features of the image based on the adjusted image; extracting noise features and texture features of the image based on the plurality of image blocks; obtaining a feature extraction result of the image by fusing the brightness features, the contrast features, the noise features, and the texture features; and tuning the image signal processor based on the feature extraction result and the current parameters of the image signal processor, thereby improving the accuracy and effectiveness of feature extraction.

[0005] Optionally, the step of extracting the brightness and contrast features of the image based on the adjusted image may include: obtaining the brightness of each color channel in a plurality of color channels of the adjusted image; and determining the brightness and contrast features of the image based on the brightness of each color channel in a plurality of color channels of the adjusted image and / or predetermined parameters.

[0006] Optionally, determining the brightness and contrast features of the image based on the brightness and / or predetermined parameters of each color channel in the adjusted image may include: determining the average brightness of the multiple color channels of the adjusted image based on the brightness of each color channel; determining the brightness features of the image based on the average brightness of the multiple color channels of the adjusted image; and determining the contrast features of the image based on the predetermined parameters, the brightness of each color channel in the adjusted image, and the average brightness of the multiple color channels of the adjusted image.

[0007] Optionally, determining the contrast characteristics of the image based on the predetermined parameters, the brightness of each color channel in the multiple color channels of the adjusted image, and the average brightness of the multiple color channels of the adjusted image may include: determining the brightness variation trend of each color channel in the multiple color channels of the adjusted image based on the predetermined parameters and the brightness of each color channel in the multiple color channels of the adjusted image; and determining the contrast characteristics of the image based on the brightness variation trend of each color channel in the multiple color channels of the adjusted image and the average brightness of the multiple color channels of the adjusted image.

[0008] Optionally, the step of extracting noise features and texture features of the image based on the plurality of image patches may include: obtaining the brightness of each color channel in the plurality of color channels of each image patch; determining the brightness gradient of each color channel in the plurality of color channels of each image patch; determining the noise features of the image based on the brightness gradient of each color channel in the plurality of image patches; and performing texture extraction on each image patch to obtain the texture features of the image.

[0009] Optionally, the step of tuning the image signal processor based on the feature extraction result and the current parameters of the image signal processor may include: determining updated parameters of the image signal processor based on the current parameters of the image signal processor and the feature extraction result; and sending the updated parameters to the image signal processor so that the image signal processor can tune its parameters based on the updated parameters, thereby improving the tuning effect.

[0010] Optionally, determining the updated parameters of the image signal processor based on the current parameters of the image signal processor and the feature extraction results may include: inputting the current parameters of the image signal processor and the feature extraction results into a parameter tuning model to obtain the updated parameters, wherein the parameter tuning model includes an action network and a comment network.

[0011] According to an exemplary embodiment of the present disclosure, a model training method is provided, comprising: in response to a training image being input into an image signal processor, acquiring current parameters of the image signal processor and a first output image; extracting features of the first output image using a feature extraction model; tuning the image signal processor based on the current parameters of the image signal processor and the features of the first output image; acquiring a second output image by inputting the training image into the tuned image signal processor; determining a loss of the feature extraction model based on the first output image and the second output image; and adjusting the parameters of the feature extraction model based on the loss, thereby improving the model training effect.

[0012] Optionally, the step of tuning the image signal processor based on the current parameters of the image signal processor and the features of the first output image may include: inputting the current parameters of the image signal processor and the features of the first output image into a tuning model to obtain updated parameters; and sending the updated parameters to the image signal processor so that the image signal processor tunes its parameters based on the updated parameters, wherein the tuning model includes an action network and a comment network.

[0013] Optionally, the step of tuning the image signal processor based on the current parameters of the image signal processor and the features of the first output image may include: determining updated parameters of the image signal processor based on the current parameters of the image signal processor and the features of the first output image; and sending the updated parameters to the image signal processor so that the image signal processor can tune its parameters based on the updated parameters.

[0014] Optionally, determining the updated parameters of the image signal processor based on the current parameters of the image signal processor and the features of the first output image may include: inputting the current parameters of the image signal processor and the features of the first output image into a parameter tuning model to obtain the updated parameters, wherein the parameter tuning model includes an action network and a comment network.

[0015] Optionally, the step of extracting features from the first output image using a feature extraction model may include: obtaining an adjusted image of the first output image by adjusting the size of the first output image, and obtaining multiple image blocks of the first output image by segmenting the first output image; extracting brightness and contrast features of the first output image based on the adjusted image; extracting noise and texture features of the first output image based on the multiple image blocks; and obtaining features of the first output image by fusing the brightness features, the contrast features, the noise features, and the texture features.

[0016] Optionally, determining the loss of the feature extraction model based on the first output image and the second output image may include: determining the difference between the quality score of the first output image and the quality score of the second output image; and determining the loss of the feature extraction model based on the difference.

[0017] According to exemplary embodiments of the present disclosure, an apparatus for parameter tuning of an image signal processor is provided, comprising: an image processing unit configured to obtain an adjusted image of the image by resizing the image, and to obtain a plurality of image blocks of the image by slicing the image, wherein the image is an output image of the image signal processor obtained in response to an input image being input to the image signal processor; a first extraction unit configured to extract brightness features and contrast features of the image based on the adjusted image; a second extraction unit configured to extract noise features and texture features of the image based on the plurality of image blocks; a feature fusion unit configured to obtain a feature extraction result of the image by fusing the brightness features, the contrast features, the noise features, and the texture features; and a parameter tuning unit configured to tune the image signal processor based on the feature extraction result and current parameters of the image signal processor.

[0018] Optionally, the first extraction unit may be configured to: acquire the brightness of each color channel in a plurality of color channels of the adjusted image; and determine the brightness features and contrast features of the image based on the brightness of each color channel in the plurality of color channels of the adjusted image and / or predetermined parameters.

[0019] Optionally, the first extraction unit may be configured to: determine the average brightness of the multiple color channels of the adjusted image based on the brightness of each color channel in the multiple color channels of the adjusted image; determine the brightness feature of the image based on the average brightness of the multiple color channels of the adjusted image; and determine the contrast feature of the image based on the predetermined parameters, the brightness of each color channel in the multiple color channels of the adjusted image, and the average brightness of the multiple color channels of the adjusted image.

[0020] Optionally, the first extraction unit may be configured to: determine the brightness variation trend of each color channel in the multiple color channels of the adjusted image based on the predetermined parameters and the brightness of each color channel in the multiple color channels of the adjusted image; and determine the contrast features of the image based on the brightness variation trend of each color channel in the multiple color channels of the adjusted image and the average brightness of the multiple color channels of the adjusted image.

[0021] Optionally, the second extraction unit may be configured to: acquire the brightness of each color channel in each of the multiple color channels of each image block in the plurality of image blocks; determine the brightness gradient of each color channel in each of the multiple color channels of each image block in the plurality of image blocks; determine the noise features of the image based on the brightness gradient of each color channel in each of the multiple color channels of each image block in the plurality of image blocks; and perform texture extraction on each of the multiple image blocks in the plurality of image blocks to obtain the texture features of the image.

[0022] Optionally, the parameter tuning unit may be configured to: determine updated parameters of the image signal processor based on the current parameters of the image signal processor and the feature extraction results; and send the updated parameters to the image signal processor so that the image signal processor performs parameter tuning based on the updated parameters.

[0023] Optionally, the parameter tuning unit can be configured to input the current parameters of the image signal processor and the feature extraction results into the parameter tuning model to obtain the updated parameters, wherein the parameter tuning model includes an action network and a comment network.

[0024] According to exemplary embodiments of the present disclosure, a model training apparatus is provided, comprising: a first acquisition unit configured to acquire current parameters of the image signal processor and a first output image in response to a training image being input into the image signal processor; a feature extraction unit configured to extract features of the first output image using a feature extraction model; a parameter tuning unit configured to tune the image signal processor based on the current parameters of the image signal processor and the features of the first output image; a second acquisition unit configured to acquire a second output image by inputting the training image into the parameter-tuned image signal processor; a loss determination unit configured to determine a loss of the feature extraction model based on the first output image and the second output image; and a parameter adjustment unit configured to adjust the parameters of the feature extraction model based on the loss.

[0025] Optionally, the parameter tuning unit may be configured to: input the current parameters of the image signal processor and the features of the first output image into the parameter tuning model to obtain updated parameters; and send the updated parameters to the image signal processor so that the image signal processor performs parameter tuning based on the updated parameters, wherein the parameter tuning model includes an action network and a comment network.

[0026] Optionally, the parameter tuning unit may be configured to: determine updated parameters of the image signal processor based on the current parameters of the image signal processor and the features of the first output image; and send the updated parameters to the image signal processor so that the image signal processor performs parameter tuning based on the updated parameters.

[0027] Optionally, the parameter tuning unit can be configured to input the current parameters of the image signal processor and the feature extraction results into the parameter tuning model to obtain the updated parameters, wherein the parameter tuning model includes an action network and a comment network.

[0028] Optionally, the feature extraction unit may be configured to: obtain an adjusted image of the first output image by adjusting the size of the first output image, and obtain multiple image blocks of the first output image by slicing the first output image into blocks; extract brightness and contrast features of the first output image based on the adjusted image; extract noise and texture features of the first output image based on the multiple image blocks; and obtain features of the first output image by fusing the brightness features, the contrast features, the noise features, and the texture features.

[0029] Optionally, the loss determination unit may be configured to: determine the difference between the quality score of the first output image and the quality score of the second output image; and determine the loss of the feature extraction model based on the difference.

[0030] According to an exemplary embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements a parameter tuning method and a model training method for an image signal processor according to an exemplary embodiment of the present disclosure.

[0031] According to an exemplary embodiment of the present disclosure, a computing device is provided, comprising: at least one processor; and at least one memory storing a computer program, wherein when the computer program is executed by the at least one processor, it implements a method for parameter tuning of an image signal processor and a model training method according to an exemplary embodiment of the present disclosure.

[0032] According to exemplary embodiments of the present disclosure, a computer program product is provided, wherein the instructions in the computer program product can be executed by a processor of a computer device to perform a method for parameter tuning of an image signal processor and a model training method according to exemplary embodiments of the present disclosure.

[0033] A method and apparatus for tuning the parameters of an image signal processor according to exemplary embodiments of the present disclosure include: obtaining an adjusted image by adjusting the size of an image; obtaining multiple image blocks of the image by slicing the image into blocks, wherein the image is an output image of the image signal processor obtained in response to an input image being input into the image signal processor; extracting brightness and contrast features of the image based on the adjusted image; extracting noise and texture features of the image based on the multiple image blocks; obtaining a feature extraction result of the image by fusing the brightness, contrast, noise, and texture features; and tuning the image signal processor based on the feature extraction result and the current parameters of the image signal processor, thereby improving the parameter tuning effect of the image signal processor by improving the accuracy and effectiveness of feature extraction.

[0034] According to exemplary embodiments of the present disclosure, a model training method and apparatus acquire current parameters and a first output image of an image signal processor in response to a training image being input into the image signal processor; extract features from the first output image using a feature extraction model; tune the image signal processor using a parameter tuning model based on the current parameters and features of the first output image; acquire a second output image by inputting the training image into the tuned image signal processor; determine the loss of the feature extraction model based on the first and second output images; and adjust the parameters of the feature extraction model based on the loss, thereby improving the model training effect.

[0035] Further aspects and / or advantages of the general concept of this disclosure will be set forth in part in the description which follows, and in part will be clear from the description or may be learned by practice of the general concept of this disclosure. Attached Figure Description

[0036] The above and other objects and features of exemplary embodiments of this disclosure will become clearer from the following description taken in conjunction with the accompanying drawings, which exemplarily illustrate the embodiments, wherein: Figure 1 A flowchart illustrating a method for tuning the parameters of an image signal processor according to an exemplary embodiment of the present disclosure is shown. Figure 2 A schematic diagram illustrating the extraction of features from an input image according to an exemplary embodiment of the present disclosure; Figure 3 A schematic diagram illustrating a method for tuning the parameters of an image signal processor according to an exemplary embodiment of the present disclosure; Figure 4 A flowchart illustrating a model training method according to an exemplary embodiment of the present disclosure is provided. Figure 5 A schematic diagram illustrating a model training method according to an exemplary embodiment of the present disclosure is shown. Figure 6 A block diagram of an apparatus for tuning an image signal processor according to an exemplary embodiment of the present disclosure is shown. Figure 7 A block diagram illustrating a model training apparatus according to exemplary embodiments of the present disclosure; and Figure 8 A schematic diagram of a computing device according to an exemplary embodiment of the present disclosure is shown. Detailed Implementation

[0037] Exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings, examples of which are illustrated in the drawings, wherein the same reference numerals always refer to the same components. The embodiments will now be described with reference to the accompanying drawings in order to explain the present disclosure.

[0038] Figure 1 A flowchart illustrating a method for tuning an image signal processor according to an exemplary embodiment of the present disclosure is shown. Figure 2 A schematic diagram illustrating the extraction of features from an input image according to an exemplary embodiment of the present disclosure. Figure 3 A schematic diagram illustrating a method for tuning the parameters of an image signal processor according to an exemplary embodiment of the present disclosure is shown. Figure 1 The method for adjusting the parameters of the image signal processor can be executed by an electronic device or the processor of an electronic device. Figure 1 The method for parameter tuning of an image signal processor is applicable to any image signal processor (ISP) that requires parameter tuning.

[0039] Reference Figure 1 In step S101, an adjusted image is obtained by adjusting the size of the image, and multiple image blocks of the image are obtained by slicing the image. Here, the image can be the output image of an image signal processor. For example, the image can be the output image of the image signal processor obtained in response to an input image being input into the image signal processor. Here, the image signal processor can be any image signal processor that requires parameter tuning, and this disclosure does not limit it.

[0040] like Figure 2 As shown, first, a scaled image and a sliced ​​image block are obtained. Here, the image size can be adjusted using pre-set scaling rules to obtain an adjusted image (e.g., Figure 2The scaled image in the image). Here, the image can be segmented using pre-set segmentation rules to obtain multiple image blocks of the image (e.g., Figure 2 Image blocks after being sliced ​​(in the image).

[0041] like Figure 2 As shown, after scaling and segmenting the image, the scaled image is sent to the global branch, and the segmented image blocks are sent to the local branch. The global branch can be understood by those skilled in the art as a processing branch that handles the overall features of the image; the local branch can be understood by those skilled in the art as a processing branch that handles the local features of the image.

[0042] As an example, referencing how humans process visual information, global branches can include brightness branches and contrast branches, while local branches can include texture branches and noise branches.

[0043] In step S102, the brightness and contrast features of the image are extracted based on the adjusted image.

[0044] As an example, brightness and contrast features are global features, meaning they require a broader focus on the entire image. Here, brightness features can represent the overall brightness of the image, and contrast features can represent the overall contrast of the image.

[0045] For example, such as Figure 2 As shown, the brightness feature extraction operator Feature extraction is performed on the adjusted image, and the dimensionality of the features extracted by the brightness feature extraction operator is reduced by average pooling to obtain the brightness features of the image. For example, average pooling can be performed using an average pooling layer. Average pooling, also known as mean pooling, is a pooling method that reduces the dimensionality of features by averaging all values ​​within a local receptive field.

[0046] For example, such as Figure 2 As shown, the contrast feature extraction operator Feature extraction is performed on the adjusted image, and the dimensionality of the features extracted by the contrast feature extraction operator is reduced by average pooling to obtain the contrast features of the image.

[0047] In an exemplary embodiment of this disclosure, extracting brightness and contrast features of the image based on the adjusted image may include: obtaining the brightness of each color channel in a plurality of color channels of the adjusted image; and determining the brightness and contrast features of the image based on the brightness of each color channel in the plurality of color channels of the adjusted image and / or predetermined parameters. Here, the plurality of color channels may be, for example, but not limited to, red (R), green (G), and blue (B) color channels.

[0048] In an exemplary embodiment of this disclosure, determining the brightness and contrast features of the image based on the brightness and / or predetermined parameters of each color channel in the plurality of color channels of the adjusted image may include: determining the average brightness of the plurality of color channels of the adjusted image based on the brightness of each color channel in the plurality of color channels; determining the brightness feature of the image based on the average brightness of the plurality of color channels of the adjusted image; and determining the contrast feature of the image based on the predetermined parameters, the brightness of each color channel in the plurality of color channels of the adjusted image, and the average brightness of the plurality of color channels of the adjusted image, thereby improving the accuracy of the brightness and contrast features. The average brightness of the plurality of color channels may be the average brightness of all color channels in the plurality of color channels. For example, if the plurality of color channels includes 5 color channels, each color channel has its own brightness, and the 5 color channels have a total of 5 brightness levels, then the average brightness of the plurality of color channels is the average of these 5 brightness levels. For example, if multiple color channels include 3 color channels, each color channel has its own brightness, and the 3 color channels have a total of 3 brightness, then the average brightness of the multiple color channels is the average of these 3 brightness.

[0049] As an example, use the following formula To extract brightness features. Here, c represents the brightness feature of the adjusted image. one, This indicates that the brightness of one of the multiple color channels in the adjusted image is adjusted. one), Indicates the R color channel. Indicates the G color channel. Indicates the B color channel. This represents the average function. This represents the average brightness of multiple color channels of the adjusted image.

[0050] In an exemplary embodiment of this disclosure, determining the contrast characteristics of an image based on the predetermined parameters, the brightness of each color channel in the plurality of color channels of the adjusted image, and the average brightness of the plurality of color channels of the adjusted image may include: determining the brightness variation trend of each color channel in the plurality of color channels of the adjusted image based on the predetermined parameters and the brightness of each color channel in the plurality of color channels of the adjusted image; and determining the contrast characteristics of the image based on the brightness variation trend of each color channel in the plurality of color channels of the adjusted image and the average brightness of the plurality of color channels of the adjusted image, thereby avoiding the influence of brightness variations and improving the accuracy of the contrast characteristics. For example, the predetermined parameters may be, but are not limited to, a Laplacian operator. Here, the brightness variation trend includes the value of the brightness variation and the direction of the brightness variation (e.g., whether the brightness increases or decreases).

[0051] As an example, use the following formula To extract contrast features. Here, c represents the contrast feature of the adjusted image. one, This indicates that the brightness of one of the multiple color channels in the adjusted image is adjusted. one), Indicates the R color channel. Indicates the G color channel. Indicates the B color channel. This represents the average brightness of multiple color channels of the adjusted image. This represents the Laplace operator.

[0052] In step S103, noise features and texture features of the image are extracted based on the plurality of image patches. Here, noise features and texture features are local features, which means that noise features and texture features need to be more focused on each image patch. Here, the noise features of the image may include the noise features of each image patch of the image, and the texture features of the image may include the texture features of each image patch of the image.

[0053] For example, such as Figure 2 As shown, the noise feature extraction operator The adjusted image is used to extract features, and the dimensionality of the features extracted by the brightness feature extraction operator is reduced by using a max pooling layer and a 1x1 convolutional layer to obtain the brightness features of the image.

[0054] For example, such as Figure 2 As shown, the texture feature extraction operator The adjusted image is used to extract features, and the dimensionality of the features extracted by the contrast feature extraction operator is reduced by using a max pooling layer and a 1x1 convolutional layer to obtain the contrast features of the image.

[0055] In an exemplary embodiment of this disclosure, extracting noise and texture features of an image based on the plurality of image patches may include: obtaining the brightness of each color channel in a plurality of color channels of each image patch; determining the brightness gradient of each color channel in a plurality of color channels of each image patch; determining the noise features of the image based on the brightness gradient of each color channel in a plurality of image patches; and performing texture extraction on each image patch to obtain the texture features of the image, thereby avoiding loss of detail and improving the accuracy of noise and texture features. Here, texture extraction can be performed on each image patch using a filter. The filter may be, for example, but not limited to, a Gabor filter. Extracting texture features using a Gabor filter can eliminate the influence of noise. Gabor filters have good noise resistance and biomimetic characteristics.

[0056] As an example, use the following formula To extract noise features from each of the multiple image patches. Here, c represents the noise features of each of the plurality of image blocks. one, This represents the brightness of one color channel in the multiple color channels of each of the multiple image blocks. one), Indicates the R color channel. Indicates the G color channel. Indicates the B color channel. This represents the gradient.

[0057] As an example, use the following formula To extract each image block from the plurality of image blocks Texture features in terms of direction. Here, Represents each of the plurality of image blocks. Texture features of direction, c represents one, This represents the brightness of one color channel from the multiple color channels (e.g., R, G, and B color channels) of each of the multiple image blocks. one), Indicates the R color channel. Indicates the G color channel. Indicates the B color channel. Indicates the direction of the texture, including 0 degrees, 45 degrees, 90 degrees, and 135 degrees. express The Gabor filter has different directions, where θ represents the different directions of the Gabor filter. For example, when θ is 0, it means that the texture features in the horizontal direction are extracted.

[0058] In step S104, the feature extraction result of the image is obtained by fusing the brightness feature, the contrast feature, the noise feature, and the texture feature. For example, as... Figure 2 As shown, the brightness features, contrast features, noise features, and texture features can be fused using a fully connected layer.

[0059] By using the feature extraction results according to exemplary embodiments of the present disclosure, the accuracy of low-level visual features of an image can be improved.

[0060] In step S105, the image signal processor is adjusted based on the feature extraction results and the current parameters of the image signal processor.

[0061] In an exemplary embodiment of this disclosure, parameter tuning of the image signal processor based on the feature extraction result and the current parameters of the image signal processor may include: determining updated parameters of the image signal processor based on the current parameters of the image signal processor and the feature extraction result; and sending the updated parameters to the image signal processor so that the image signal processor performs parameter tuning based on the updated parameters. That is, the image signal processor adjusts its own parameters according to the updated parameters. For example, the image signal processor adjusts its own parameters (the parameters of the image signal processor) to the updated parameters.

[0062] In an exemplary embodiment of this disclosure, determining the updated parameters of the image signal processor based on the current parameters of the image signal processor and the feature extraction results may include: inputting the current parameters of the image signal processor and the feature extraction results into a parameter tuning model to obtain the updated parameters.

[0063] In an exemplary embodiment of this disclosure, the parameter tuning model may include an actor network and a critical network.

[0064] As an example, the hyperparameter tuning model described herein can be designed based on the Soft Actor-Critic (SAC) algorithm. The Soft Actor-Critic (SAC) algorithm is a deep learning-based reinforcement learning algorithm that combines maximum entropy reinforcement learning and the actor-critic framework; it is an off-policy approach. By introducing the maximum entropy reinforcement learning framework, the SAC algorithm automatically adjusts the exploration level during policy optimization, effectively solving the problems of exploration-exploitation tradeoffs and training stability in traditional deep reinforcement learning algorithms.

[0065] As an example, in this disclosure, the hyperparameter tuning model may include an Actor network and a Critical network, wherein the Actor network may be a Policy network and the Critical network may be a Twined-Q network. The Policy network can generate new ISP parameters based on the state. The Twined-Q network can evaluate the state and actions to aid in training the hyperparameter tuning model. The Actor and Critical networks are trained adversarially and eventually converge.

[0066] As an example, such as Figure 3 As shown, the input image is fed into an image signal processor (ISP) to obtain an output image. A feature extractor extracts features from the output image to obtain its features. These features, along with the ISP parameters, are stored in a data buffer. Then, these features and ISP parameters are fed into an action network and a comment network to obtain updated parameters. These updated parameters are then fed back into the ISP to tune its parameters.

[0067] Figure 4 A flowchart illustrating a model training method according to an exemplary embodiment of the present disclosure is shown. Figure 5 A schematic diagram illustrating a model training method according to an exemplary embodiment of the present disclosure is shown. Figure 4 The model training methods described above can be applied to the training of feature extraction models.

[0068] Reference Figure 4 In step S401, in response to the training image being input into the image signal processor, the current parameters of the image signal processor and the first output image are obtained. Here, the image signal processor can be any image signal processor.

[0069] In step S402, features of the first output image are extracted using a feature extraction model.

[0070] In an exemplary embodiment of this disclosure, extracting features of a first output image using a feature extraction model may include: obtaining an adjusted image of the first output image by adjusting the size of the first output image, and obtaining multiple image blocks of the first output image by segmenting the first output image; extracting brightness and contrast features of the first output image based on the adjusted image; extracting noise and texture features of the first output image based on the multiple image blocks; and obtaining features of the first output image by fusing the brightness features, the contrast features, the noise features, and the texture features.

[0071] In an exemplary embodiment of this disclosure, extracting brightness and contrast features of a first output image based on the adjusted image may include: obtaining the brightness of each color channel among a plurality of color channels of the adjusted image; and determining the brightness and contrast features of the first output image based on the brightness of each color channel among the plurality of color channels of the adjusted image and / or predetermined parameters. Here, the plurality of color channels may be, for example, but not limited to, red (R), green (G), and blue (B) color channels.

[0072] In an exemplary embodiment of this disclosure, determining the brightness and contrast features of a first output image based on the brightness and / or predetermined parameters of each color channel in the plurality of color channels of the adjusted image may include: determining the average brightness of the plurality of color channels of the adjusted image based on the brightness of each color channel in the plurality of color channels; determining the brightness features of the first output image based on the average brightness of the plurality of color channels of the adjusted image; and determining the contrast features of the first output image based on the predetermined parameters, the brightness of each color channel in the plurality of color channels of the adjusted image, and the average brightness of the plurality of color channels of the adjusted image.

[0073] In an exemplary embodiment of this disclosure, determining the contrast characteristics of a first output image based on the predetermined parameters, the brightness of each color channel in the plurality of color channels of the adjusted image, and the average brightness of the plurality of color channels of the adjusted image may include: determining the brightness variation trend of each color channel in the plurality of color channels of the adjusted image based on the predetermined parameters and the brightness of each color channel in the plurality of color channels of the adjusted image; and determining the contrast characteristics of the first output image based on the brightness variation trend of each color channel in the plurality of color channels of the adjusted image and the average brightness of the plurality of color channels of the adjusted image.

[0074] In an exemplary embodiment of this disclosure, extracting noise features and texture features of a first output image based on the plurality of image blocks may include: obtaining the brightness of each color channel in a plurality of color channels of each image block in the plurality of image blocks; determining the brightness gradient of each color channel in a plurality of color channels of each image block in the plurality of image blocks; determining the noise features of the first output image based on the brightness gradient of each color channel in a plurality of color channels of each image block in the plurality of image blocks; and performing texture extraction on each image block in the plurality of image blocks to obtain the texture features of the first output image.

[0075] In step S403, the image signal processor is adjusted based on its current parameters and the features of the first output image.

[0076] In an exemplary embodiment of this disclosure, parameter tuning of the image signal processor based on its current parameters and features of the first output image may include: inputting the current parameters of the image signal processor and features of the first output image into a parameter tuning model to obtain updated parameters; and sending the updated parameters to the image signal processor so that the image signal processor performs parameter tuning based on the updated parameters, wherein the parameter tuning model includes an action network and a comment network.

[0077] In an exemplary embodiment of this disclosure, adjusting the parameters of the image signal processor based on its current parameters and the features of the first output image may include: determining updated parameters of the image signal processor based on its current parameters and the features of the first output image; and sending the updated parameters to the image signal processor so that the image signal processor adjusts its parameters based on the updated parameters.

[0078] In an exemplary embodiment of this disclosure, determining the updated parameters of the image signal processor based on the current parameters of the image signal processor and the features of the first output image may include: inputting the current parameters of the image signal processor and the features of the first output image into a parameter tuning model to obtain the updated parameters.

[0079] In an exemplary embodiment of this disclosure, the parameter tuning model includes an action network and a comment network.

[0080] As an example, the hyperparameter tuning model described herein can be designed based on the Soft Actor-Critic (SAC) algorithm. The Soft Actor-Critic (SAC) algorithm is a deep learning-based reinforcement learning algorithm that combines maximum entropy reinforcement learning and the actor-critic framework; it is an off-policy approach. By introducing the maximum entropy reinforcement learning framework, the SAC algorithm automatically adjusts the exploration level during policy optimization, effectively solving the problems of exploration-exploitation tradeoffs and training stability in traditional deep reinforcement learning algorithms.

[0081] As an example, in this disclosure, the hyperparameter tuning model may include an Actor network and a Critical network, wherein the Actor network may be a Policy network and the Critical network may be a Twined-Q network. The Policy network can generate new ISP parameters based on the state. The Twined-Q network can evaluate the state and actions to aid in training the hyperparameter tuning model. The Actor and Critical networks are trained adversarially and eventually converge.

[0082] As an example, this disclosure incorporates an Image Quality Assessment (IQA) model to evaluate image quality and output an image-related score. Since the IQA score is always positive, it affects the Twined-Q network's evaluation of actions in the Actor network within the parameter tuning model. In this disclosure, the difference in IQA scores between the states at time t and time t+1 is calculated as the reward for the Twined-Q network when evaluating actions in the Actor network.

[0083] In step S404, the second output image is obtained by inputting the training image into the parameter-tuned image signal processor.

[0084] In step S405, the loss used by the feature extraction model is determined based on the first output image and the second output image.

[0085] In an exemplary embodiment of this disclosure, determining the loss of the feature extraction model based on a first output image and a second output image may include: determining the difference between the quality scores of the first output image and the second output image; and determining the loss of the feature extraction model based on the difference.

[0086] As an example, it can be done through the formula To calculate the loss of the feature extraction model. Here, Let b represent the loss of the feature extraction model, and b represent the noise features and texture features. f This represents the output of a fully connected layer, indicating the noise or texture features. express t The output of the fully connected layer for noise or texture features at time +1. express t The noise or texture features at any given time are the output of the fully connected layer. express t The quality score of the output image at time +1 (e.g., the second output image). express t The quality score of the output image at time (e.g., the first output image), where ∑ represents summation.

[0087] In step S406, the parameters of the feature extraction model are adjusted based on the loss.

[0088] As an example, such as Figure 5 As shown, in t At a certain time, the input image is fed into the image signal processor (ISP) to obtain the first output image. The feature extractor then extracts features from the first output image to obtain its features. t The features of the first output image at each time step and the parameters of the Image Signal Processor (ISP) are stored together in a data buffer. Then, these features and ISP parameters are input into the action and comment networks of the hyperparameter tuning model to obtain updated parameters. These updated parameters are then input into the ISP, thereby tuning the ISP based on these updated parameters. t At time +1, the input image is again input into the image signal processor (ISP) to obtain the second output image. The first and second output images are then input into an image quality assessment model to evaluate image quality, resulting in quality scores for the first and second output images. The difference between the quality scores of the second and first output images can be used as the reward for the parameter tuning model and the loss for the feature extractor.

[0089] The feature extraction model according to exemplary embodiments of this disclosure can be deployed in robots or factory monitors to perform machine vision tasks such as object detection and image segmentation. High-level visual information is supplemented by combining low-level visual features extracted by the feature extraction model of exemplary embodiments of this disclosure with semantic features.

[0090] The feature extraction model according to exemplary embodiments of this disclosure can extract low-level visual features to improve the performance of large language models (LLMs) on low-level tasks such as certain IQA tasks.

[0091] The method for tuning the parameters of an image signal processor according to exemplary embodiments of the present disclosure can be deployed in a software or hardware image signal processor (ISP) pipeline. By using the method for tuning the parameters of an image signal processor according to exemplary embodiments of the present disclosure, parameter adjustment of commercial image signal processors (ISPs) can be achieved, thereby improving the image processing performance of the image signal processor (ISP).

[0092] The above has been combined Figures 1 to 5 A method for parameter tuning of an image signal processor and a model training method according to exemplary embodiments of the present disclosure have been described. Hereinafter, reference will be made to... Figure 6 and Figure 7 An apparatus and its units for tuning the parameters of an image signal processor, and a model training apparatus and its units, according to exemplary embodiments of the present disclosure, will be described.

[0093] Figure 6 A block diagram of an apparatus for tuning an image signal processor according to an exemplary embodiment of the present disclosure is shown.

[0094] Reference Figure 6 The apparatus for adjusting the parameters of an image signal processor includes an image processing unit 61, a first extraction unit 62, a second extraction unit 63, a feature fusion unit 64, and a parameter adjustment unit 65. The image processing unit 61 is configured to obtain an adjusted image of the image by adjusting the size of the image, and to obtain a plurality of image blocks of the image by slicing the image, wherein the image is an output image of the image signal processor obtained in response to an input image being input to the image signal processor.

[0095] The first extraction unit 62 is configured to extract the brightness and contrast features of the image based on the adjusted image.

[0096] In an exemplary embodiment of this disclosure, the first extraction unit 62 may be configured to: acquire the brightness of each of the plurality of color channels of the adjusted image; and determine the brightness features and contrast features of the image based on the brightness of each of the plurality of color channels of the adjusted image and / or predetermined parameters. Here, the plurality of color channels may be, for example, but not limited to, red (R), green (G), and blue (B) color channels.

[0097] In an exemplary embodiment of this disclosure, the first extraction unit 62 may be configured to: determine the average brightness of the plurality of color channels of the adjusted image based on the brightness of each color channel of the adjusted image; determine the brightness feature of the image based on the average brightness of the plurality of color channels of the adjusted image; and determine the contrast feature of the image based on the predetermined parameters, the brightness of each color channel of the adjusted image, and the average brightness of the plurality of color channels of the adjusted image.

[0098] In an exemplary embodiment of this disclosure, the first extraction unit 62 may be configured to: determine the brightness variation trend of each color channel in the plurality of color channels of the adjusted image based on the predetermined parameters and the brightness of each color channel in the plurality of color channels of the adjusted image; and determine the contrast features of the image based on the brightness variation trend of each color channel in the plurality of color channels of the adjusted image and the average brightness of the plurality of color channels of the adjusted image.

[0099] The second extraction unit 63 is configured to extract noise features and texture features of the image based on the plurality of image blocks.

[0100] In an exemplary embodiment of this disclosure, the second extraction unit 63 may be configured to: acquire the brightness of each color channel in a plurality of color channels of each image block in the plurality of image blocks; determine the brightness gradient of each color channel in a plurality of color channels of each image block in the plurality of image blocks; determine the noise features of the image based on the brightness gradient of each color channel in a plurality of color channels of each image block in the plurality of image blocks; and perform texture extraction on each image block in the plurality of image blocks to obtain the texture features of the image.

[0101] The feature fusion unit 64 is configured to obtain the feature extraction result of the image by fusing the brightness feature, the contrast feature, the noise feature and the texture feature.

[0102] The parameter tuning unit 65 is configured to tune the image signal processor based on the feature extraction results and the current parameters of the image signal processor.

[0103] In an exemplary embodiment of this disclosure, the parameter tuning unit 65 may be configured to determine updated parameters of the image signal processor based on the current parameters of the image signal processor and the feature extraction results; and to send the updated parameters to the image signal processor so that the image signal processor performs parameter tuning based on the updated parameters.

[0104] In an exemplary embodiment of this disclosure, the parameter tuning unit 65 may be configured to input the current parameters of the image signal processor and the feature extraction results into the parameter tuning model to obtain the updated parameters, wherein the parameter tuning model includes an action network and a comment network.

[0105] Figure 7 A block diagram of a model training apparatus according to an exemplary embodiment of the present disclosure is shown.

[0106] Reference Figure 7 The model training device includes a first acquisition unit 71, a feature extraction unit 72, a processor parameter tuning unit 73, a second acquisition unit 74, a loss determination unit 75, and a parameter adjustment unit 76.

[0107] The first acquisition unit 71 is configured to acquire the current parameters of the image signal processor and the first output image in response to the training image being input into the image signal processor.

[0108] The feature extraction unit 72 is configured to extract features of the first output image through a feature extraction model.

[0109] In an exemplary embodiment of this disclosure, the feature extraction unit 72 may be configured to: obtain an adjusted image of the first output image by adjusting the size of the first output image, and obtain a plurality of image blocks of the first output image by performing block processing on the first output image; extract brightness features and contrast features of the first output image based on the adjusted image; extract noise features and texture features of the first output image based on the plurality of image blocks; and obtain features of the first output image by fusing the brightness features, the contrast features, the noise features and the texture features.

[0110] In an exemplary embodiment of this disclosure, the feature extraction unit 72 may be configured to: acquire the brightness of each of the plurality of color channels of the adjusted image; and determine the brightness features and contrast features of the first output image based on the brightness of each of the plurality of color channels of the adjusted image and / or predetermined parameters.

[0111] In an exemplary embodiment of this disclosure, the feature extraction unit 72 may be configured to: determine the average brightness of the plurality of color channels of the adjusted image based on the brightness of each color channel of the adjusted image; determine the brightness feature of a first output image based on the average brightness of the plurality of color channels of the adjusted image; and determine the contrast feature of the first output image based on the predetermined parameters, the brightness of each color channel of the adjusted image, and the average brightness of the plurality of color channels of the adjusted image.

[0112] In an exemplary embodiment of this disclosure, the feature extraction unit 72 may be configured to: determine the brightness variation trend of each color channel in the plurality of color channels of the adjusted image based on the predetermined parameters and the brightness of each color channel in the plurality of color channels of the adjusted image; and determine the contrast features of the first output image based on the brightness variation trend of each color channel in the plurality of color channels of the adjusted image and the average brightness of the plurality of color channels of the adjusted image.

[0113] In an exemplary embodiment of this disclosure, the feature extraction unit 72 may be configured to: acquire the brightness of each color channel in a plurality of color channels of each image block in the plurality of image blocks; determine the brightness gradient of each color channel in a plurality of color channels of each image block in the plurality of image blocks; determine the noise features of a first output image based on the brightness gradient of each color channel in a plurality of color channels of each image block in the plurality of image blocks; and perform texture extraction on each image block in the plurality of image blocks to obtain the texture features of the first output image.

[0114] The processor tuning unit 73 is configured to tune the image signal processor based on the current parameters of the image signal processor and the features of the first output image.

[0115] In an exemplary embodiment of this disclosure, the processor parameter tuning unit 73 may be configured to: input the current parameters of the image signal processor and the features of the first output image into the parameter tuning model to obtain updated parameters; and send the updated parameters to the image signal processor so that the image signal processor performs parameter tuning based on the updated parameters, wherein the parameter tuning model includes an action network and a comment network.

[0116] In an exemplary embodiment of this disclosure, the processor parameter tuning unit 73 may be configured to: determine updated parameters of the image signal processor based on the current parameters of the image signal processor and the features of the first output image; and send the updated parameters to the image signal processor so that the image signal processor performs parameter tuning based on the updated parameters.

[0117] In an exemplary embodiment of this disclosure, the processor tuning unit 73 may be configured to: input the current parameters of the image signal processor and the features of the first output image into the tuning model to obtain the updated parameters, wherein the tuning model includes an action network and a comment network.

[0118] The second acquisition unit 74 is configured to acquire a second output image by inputting the training image into a parameter-tuned image signal processor.

[0119] The loss determination unit 75 is configured to determine the loss of the feature extraction model based on the first output image and the second output image.

[0120] The parameter adjustment unit 76 is configured to adjust the parameters of the feature extraction model based on the loss.

[0121] Furthermore, according to exemplary embodiments of the present disclosure, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed, implements a method for parameter tuning of an image signal processor or a model training method according to exemplary embodiments of the present disclosure.

[0122] In an exemplary embodiment of this disclosure, the computer-readable storage medium may carry one or more programs that, when executed, perform the following steps: obtaining an adjusted image by resizing an image, and obtaining multiple image blocks of the image by slicing the image, wherein the image is an output image of the image signal processor obtained in response to an input image being input to the image signal processor; extracting brightness and contrast features of the image based on the adjusted image; extracting noise and texture features of the image based on the multiple image blocks; obtaining a feature extraction result of the image by fusing the brightness, contrast, noise, and texture features; and tuning the image signal processor based on the feature extraction result and the current parameters of the image signal processor, thereby improving the tuning effect of the image signal processor by improving the accuracy and effectiveness of feature extraction.

[0123] In an exemplary embodiment of this disclosure, the computer-readable storage medium may carry one or more programs that, when executed, perform the following steps: in response to a training image being input into an image signal processor, acquiring the current parameters of the image signal processor and a first output image; extracting features from the first output image using a feature extraction model; tuning the image signal processor using a parameter tuning model based on the current parameters of the image signal processor and the features of the first output image; acquiring a second output image by inputting the training image into the parameter-tuned image signal processor; determining the loss of the feature extraction model based on the first and second output images; and adjusting the parameters of the feature extraction model based on the loss, thereby improving the model training effect.

[0124] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In embodiments of this disclosure, a computer-readable storage medium can be any tangible medium that contains or stores a computer program that can be used by or in conjunction with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable storage medium can be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof. A computer-readable storage medium can be included in any apparatus; it can also exist independently without being assembled into that apparatus.

[0125] Furthermore, according to exemplary embodiments of the present disclosure, a computer program product is also provided, wherein the instructions in the computer program product can be executed by a processor of a computer device to perform a method for parameter tuning of an image signal processor and a model training method according to exemplary embodiments of the present disclosure.

[0126] The above has been combined Figure 6 and Figure 7 An apparatus for tuning the parameters of an image signal processor and a model training apparatus according to exemplary embodiments of the present disclosure have been described. Next, in conjunction with... Figure 8 A computing device according to exemplary embodiments of the present disclosure will be described.

[0127] Figure 8 A schematic diagram of a computing device according to an exemplary embodiment of the present disclosure is shown.

[0128] Reference Figure 8 The computing device 8 according to an exemplary embodiment of the present disclosure includes a memory 801 and a processor 802. The memory 801 stores a computer program, which, when executed by the processor 802, implements a method for tuning parameters of an image signal processor and a model training method according to an exemplary embodiment of the present disclosure.

[0129] In an exemplary embodiment of this disclosure, when the computer program is executed by the processor 802, the following steps can be implemented: obtaining an adjusted image by adjusting the size of an image, and obtaining multiple image blocks of the image by slicing the image, wherein the image is an output image of the image signal processor obtained in response to an input image being input into the image signal processor; extracting brightness and contrast features of the image based on the adjusted image; extracting noise and texture features of the image based on the multiple image blocks; obtaining a feature extraction result of the image by fusing the brightness features, the contrast features, the noise features, and the texture features; and tuning the image signal processor based on the feature extraction result and the current parameters of the image signal processor, thereby improving the parameter tuning effect of the image signal processor by improving the accuracy and effectiveness of feature extraction.

[0130] In an exemplary embodiment of this disclosure, when the computer program is executed by the processor 802, the following steps can be implemented: in response to a training image being input into an image signal processor, obtaining the current parameters of the image signal processor and a first output image; extracting features of the first output image using a feature extraction model; tuning the image signal processor using a parameter tuning model based on the current parameters of the image signal processor and the features of the first output image; obtaining a second output image by inputting the training image into the parameter-tuned image signal processor; determining the loss of the feature extraction model based on the first output image and the second output image; and adjusting the parameters of the feature extraction model based on the loss, thereby improving the model training effect.

[0131] The computing device in this disclosure may include, but is not limited to, devices such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 8 The computing device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments disclosed herein.

[0132] The above has been referred to Figures 1 to 8 Methods and apparatus for parameter tuning of an image signal processor, and model training methods and apparatus according to exemplary embodiments of the present disclosure are described. However, it should be understood that: Figure 6 and Figure 7 The apparatus for tuning the image signal processor, the model training apparatus, and their units shown can be configured as software, hardware, firmware, or any combination thereof to perform specific functions. Figure 8 The computing device shown is not limited to the components shown above, but some components may be added or removed as needed, and the above components may also be combined.

[0133] A method and apparatus for tuning the parameters of an image signal processor according to exemplary embodiments of the present disclosure obtain an adjusted image by adjusting the size of an image, and obtain multiple image blocks of the image by slicing the image into blocks, wherein the image is an output image of the image signal processor obtained in response to an input image being input into the image signal processor. Brightness and contrast features of the image are extracted based on the adjusted image, noise and texture features of the image are extracted based on the multiple image blocks, and feature extraction results of the image are obtained by fusing the brightness, contrast, noise, and texture features. The image signal processor is tuned based on the feature extraction results and the current parameters of the image signal processor, thereby improving the parameter tuning effect of the image signal processor by improving the accuracy and effectiveness of feature extraction.

[0134] According to exemplary embodiments of the present disclosure, a model training method and apparatus acquire current parameters and a first output image of an image signal processor in response to a training image being input into the image signal processor; extract features from the first output image using a feature extraction model; tune the image signal processor using a parameter tuning model based on the current parameters and features of the first output image; acquire a second output image by inputting the training image into the tuned image signal processor; determine the loss of the feature extraction model based on the first and second output images; and adjust the parameters of the feature extraction model based on the loss, thereby improving the model training effect.

[0135] Although this disclosure has been specifically shown and described with reference to exemplary embodiments thereof, those skilled in the art should understand that various changes in form and detail may be made therein without departing from the spirit and scope of this disclosure as defined by the claims.

Claims

1. A method for parameter tuning of an image signal processor, comprising: An adjusted image is obtained by resizing the image, and multiple image blocks of the image are obtained by slicing the image into blocks, wherein the image is the output image of the image signal processor obtained in response to an input image being input into the image signal processor; Based on the adjusted image, extract the brightness and contrast features of the image; Based on the multiple image patches, noise and texture features of the image are extracted; and The feature extraction result of the image is obtained by fusing the brightness feature, the contrast feature, the noise feature, and the texture feature; The image signal processor is tuned based on the feature extraction results and the current parameters of the image signal processor.

2. The method according to claim 1, wherein, The step of extracting the brightness and contrast features of the adjusted image includes: Obtain the brightness of each color channel in the multiple color channels of the adjusted image; The brightness and contrast characteristics of the image are determined based on the brightness and / or predetermined parameters of each color channel in the multiple color channels of the adjusted image.

3. The method according to claim 2, wherein, Determining the brightness and contrast features of the image based on the brightness and / or predetermined parameters of each color channel in the adjusted image includes: Based on the brightness of each color channel in the adjusted image, the average brightness of the multiple color channels of the adjusted image is determined. The brightness characteristics of the image are determined based on the average brightness of multiple color channels of the adjusted image. The contrast characteristics of the image are determined based on the predetermined parameters, the brightness of each color channel in the multiple color channels of the adjusted image, and the average brightness of the multiple color channels of the adjusted image.

4. The method according to claim 3, wherein, Determining the contrast characteristics of the image based on the predetermined parameters, the brightness of each color channel in the multiple color channels of the adjusted image, and the average brightness of the multiple color channels of the adjusted image includes: Based on the predetermined parameters and the brightness of each color channel in the multiple color channels of the adjusted image, determine the trend of brightness variation of each color channel in the multiple color channels of the adjusted image; The contrast characteristics of the image are determined based on the brightness variation trend of each color channel in the multiple color channels of the adjusted image and the average brightness of the multiple color channels of the adjusted image.

5. The method according to claim 1, wherein, The step of extracting noise and texture features of the image based on the multiple image patches includes: Obtain the brightness of each color channel in the multiple color channels of each image block in the plurality of image blocks; Determine the brightness gradient of each color channel in the multiple color channels of each of the multiple image blocks; The noise features of the image are determined based on the brightness gradient of each color channel in each of the multiple color channels of each of the multiple image blocks; Texture extraction is performed on each of the plurality of image blocks to obtain the texture features of the image.

6. The method according to claim 1, wherein, The step of adjusting the parameters of the image signal processor based on the feature extraction results and the current parameters of the image signal processor includes: Based on the current parameters of the image signal processor and the feature extraction results, the updated parameters of the image signal processor are determined; and The updated parameters are sent to the image signal processor so that the image signal processor can perform parameter tuning based on the updated parameters.

7. The method according to claim 6, wherein, The step of determining the updated parameters of the image signal processor based on the current parameters of the image signal processor and the feature extraction results includes: The current parameters of the image signal processor and the feature extraction results are input into the parameter tuning model to obtain the updated parameters. The parameter tuning model includes an action network and a comment network.

8. An apparatus for adjusting parameters of an image signal processor, comprising: An image processing unit is configured to obtain an adjusted image of the image by resizing the image, and to obtain a plurality of image blocks of the image by slicing the image, wherein the image is an output image of the image signal processor obtained in response to an input image being input to the image signal processor; The first extraction unit is configured to extract the brightness and contrast features of the image based on the adjusted image; The second extraction unit is configured to extract noise features and texture features of the image based on the plurality of image blocks; The feature fusion unit is configured to obtain the feature extraction result of the image by fusing the brightness feature, the contrast feature, the noise feature, and the texture feature; and The parameter tuning unit is configured to tune the image signal processor based on the feature extraction results and the current parameters of the image signal processor.

9. A computer-readable storage medium storing a computer program, wherein, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. A computing device, comprising: At least one processor; At least one memory storing a computer program that, when executed by the at least one processor, implements the method of any one of claims 1 to 7.

Images