| CPC G06T 5/73 (2024.01) [G06T 5/20 (2013.01); G06T 5/70 (2024.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] | 20 Claims |
|
1. An image optimization method, performed by a computing device, the method comprising:
obtaining a to-be-optimized image;
aligning the to-be-optimized image according to a standard position template including a point distribution of each object in a specific region to obtain a to-be-optimized aligned image, the to-be-optimized aligned image including a target region having points of objects that are distributed in a standard position; and
using the to-be-optimized aligned image as an input to a generation network;
performing feature extraction on the to-be-optimized aligned image using the generation network, to obtain an optimized image, wherein:
the generation network is obtained by training a to-be-trained generative adversarial deep neural network model by:
obtaining a plurality of target images:
aligning the plurality of target images respectively to obtain a plurality of aligned target images;
performing image processing on the plurality of aligned target images, to obtain a plurality of low-quality images;
generate a plurality of low-quality image pairs from the plurality of target images and the plurality of target images, each low-quality image pair including a target image and a low-quality image corresponding to the target image;
inputting the low-quality image in each low-quality image pair to a generation network in the to-be-trained generative adversarial deep neural network model, to obtain a generated image;
using the generated image and the target image in the low-quality image pair as inputs to a post-processing network in the to-be-trained generative adversarial deep neural network model;
processing the generated image and the target image in the low-quality image pair through the post-processing network, to construct a joint loss function; and
optimizing a plurality of parameters of the to-be-trained generative adversarial deep neural network model according to the joint loss function, to obtain the generation network.
|