Method of image processing based on plurality of frames of images, electronic device, and storage medium

1. A method of image processing based on a plurality of frames of images, comprising:

capturing a preview image of a scene;

capturing number (n) frames of original images of the scene, wherein a value of n for the n frames of original images is determined based on quality of the preview image;

obtaining a high dynamic range (HDR) image by performing image synthesis on the n frames of original images; and

performing artificial intelligent (AI)-based denoising on the HDR image to obtain a target denoised image,

wherein capturing the n frames of original images comprises:

determining the n frames of original images having a reference exposure amount to be captured based on the quality of the preview image, wherein the n is a natural number greater than or equal to 2;

capturing the n frames of original images meeting the reference exposure amount; and

capturing at least one frame of an original image having an exposure amount lower than the reference exposure amount.

2. The method according to claim 1 , wherein performing the AI-based denoising on the HDR image to obtain the target denoised image comprises:

performing, by a neural network model, noise characteristics identification on the HDR image, wherein the neural network model learns and obtains a mapping relation between a light sensitivity and the noise characteristics of the HDR image; and

performing denoising on the HDR image based on the identified noise characteristics to obtain the target denoised image.

3. The method according to claim 2 , wherein the neural network model is trained by taking a plurality of sample images with a plurality of light sensitivities as samples, and the training of the neural network model is completed when the noise characteristics identified by the neural network model matches with the noise characteristics labeled in a corresponding sample image.

4. The method according to claim 1 , wherein

the n frames of original images comprise at least two frames of first images having a same exposure amount and at least one frame of a second image having an exposure amount lower than the exposure amount of the first images; and

obtaining the HDR image by performing image synthesis on the n frames of images comprises:

performing multi-frame denoising on the at least two frames of the first images to obtain a synthesized denoised image; and

performing high dynamic synthesis on the synthesized denoised image and the at least one frame of the second image to obtain the HDR image.

5. The method according to claim 1 , wherein obtaining the HDR image by performing the image synthesis on the n frames of original images comprises:

inputting the n frames of original images into a high dynamic synthesis model to obtain a synthesis weight of each area of a corresponding original image; and

performing image synthesis on each area of the n frames of original images based on the synthesis weight of each area to obtain the HDR image.

6. The method according to claim 5 , wherein the high dynamic synthesis model learns and obtains a mapping relation between characteristics of each area of the n frames of original images and the synthesis weight; and the characteristics of each area is arranged to indicate an exposure amount of each area and a luminance level of a corresponding area.

7. The method according to claim 1 , wherein capturing the n frames of original images meeting the reference exposure amount comprises:

determining the reference exposure amount based on a luminance level of a scene of image capturing;

determining reference exposure duration based on the reference exposure amount and a preset reference light sensitivity; and

capturing the n frames of original images based on the reference exposure duration and the reference light sensitivity.

8. The method according to claim 7 , wherein capturing the at least one frame of the original image having the exposure amount lower than the reference exposure amount comprises:

performing compensation on the reference exposure duration based on a preset exposure compensation level to obtain compensated exposure duration shorter than the reference exposure duration; and

capturing the at least one frame of the original image based on the compensated exposure duration and the reference light sensitivity.

9. The method according to claim 7 , wherein the method further comprises:

before determining the reference exposure duration based on the reference exposure amount and the preset reference light sensitivity, setting the reference light sensitivity based on a degree of smear of the preview image or a degree of shaking of an image sensor capturing the preview image.

10. The method according to claim 7 , wherein a value of the reference light sensitivity is within a range of 100ISO to 200ISO.

11. The method according to claim 1 , wherein

the at least one frame of the original image comprises two frames of original images; and

the two frames of original images correspond to different exposure compensation levels, and the exposure compensation levels corresponding to the two frames of original images are less than EVO.

12. The method according to claim 1 , wherein an exposure compensation level corresponding to the at least one frame of the original image is within a range of EV-5 to EV-1.

13. The method according to claim 1 , wherein the n is equal to 3 or 4.

14. The method according to claim 1 , wherein the method further comprises:

after performing the AI-based denoising on the HDR image to obtain the target denoised image, converting a format of the target denoised image into a YUV format.

15. An electronic device, comprising: an image sensor, a non-transitory memory, a processor, and a computer program stored in the non-transitory memory and run by the processor, wherein

the processor comprises an image signal processing (ISP) pipeline and a graphics processing unit (GPU) connected to the ISP pipeline;

the ISP pipeline is arranged to control the image sensor to capture a preview image of a scene, capture number (n) frames of original images of the scene, and perform high dynamic synthesis on the n frames of original images to obtain a HDR image, wherein a value of n for the n frames of original images is determined based on quality of the preview image;

the GPU is arranged to perform AI-based denoising on the HDR image to obtain a target denoised image; and

capturing the n frames of original images comprises:

determining the n frames of original images having a reference exposure amount to be captured based on the quality of the preview image, wherein the n is a natural number greater than or equal to 2;

capturing the n frames of original images meeting the reference exposure amount; and

capturing at least one frame of an original image having an exposure amount lower than the reference exposure amount.

16. The electronic device according to claim 15 , wherein the GPU is further arranged to perform noise characteristics identification on the HDR image and perform denoising on the HDR image based on the identified noise characteristics to obtain the target denoised image.

17. The electronic device according to claim 15 , wherein

the n frames of original images comprise at least two frames of first images having a same exposure amount and at least one frame of a second image having an exposure amount lower than the exposure amount of the at least two frame of first images; and

the ISP pipeline is further arranged to perform multi-frame denoising on the at least two frames of first images to obtain a synthesized denoised image and perform high dynamic synthesis on the synthesized denoised image and the at least one frame of the second image to obtain the HDR image.

18. The electronic device according to claim 15 , wherein the ISP pipeline is further arranged to input the n frames of original images into a high dynamic synthesis model to obtain a synthesis weight of each area of a corresponding original image and perform image synthesis on each area of the n frames of original images based on the synthesis weight of each area to obtain the HDR image.

19. A computer-readable non-transitory storage medium, comprising a computer program stored in, wherein the computer program is capable of being executed by a processor to perform operations of:

capturing a preview image of a scene;

capturing number (n) frames of original images of the scene, wherein a value of n for the n frames of original images is determined based on quality of the preview image;

obtaining a high dynamic range (HDR) image by performing image synthesis on the n frames of original images; and

performing artificial intelligent (AI)-based denoising on the HDR image to obtain a target denoised image,

wherein capturing the n frames of original images comprises:

determining the n frames of original images having a reference exposure amount to be captured based on the quality of the preview image, wherein the n is a natural number greater than or equal to 2;

capturing the n frames of original images meeting the reference exposure amount; and

capturing at least one frame of an original image having an exposure amount lower than the reference exposure amount.