Face image restoration method, system, storage medium and device
OwnerDengyin Zhang, Weidan Yan +4inventors
InventorsDengyin Zhang, Weidan Yan, Chao Zhou, Junhao Ying, Yingying Feng, Yi Liu
Life of this patent
Estimating the term…
Abstract
The present application discloses a face image restoration method, a system, a storage medium and a device, the restoration model adopted in the above method starts from structured information of a face, generates a structured face graph based on features of the to-be-restored face image, and restores face image by the structured face graph and the decoder, which can solve the problem that it is difficult to capture and learn structured information based on the traditional convolution operation, and improve the indicators of the face restoration and enriches the visualization effect.
Claims7
A face image restoration method, comprising:
- obtaining a to-be-restored face image;
- and
- inputting the to-be-restored face image into a pre-trained restoration model to obtain a restored face image;
- wherein the restoration model comprises an encoder, a structured face graph generation module, an adaptive adjacency matrix generator, a face graph convolution network, and a decoder;
- the encoder is configured for generating face image features based on the to-be-restored face image;
- the adaptive adjacency matrix generator is configured for generating an adjacency matrix that represents connection relationships of face image feature blocks based on the face image features;
- a structured face graph generation module is configured for generating a structured face graph by using the face image feature blocks as nodes and the adjacency matrix as the node connection relationship;
- a face graph convolution network is configured for generating a restored face feature graph based on the structured face graph;
- and
- a decoder is configured for generating the restored face image based on the restored face feature graph;
- wherein the encoder comprises six residual convolution modules and five 2-fold down-sampling modules, and the six residual convolution modules are alternately connected in series with the five 2-fold down-sampling modules in the encoder;
- and
- the decoder comprises six residual convolution modules and five 2-fold up-sampling modules, and the six residual convolution modules are alternately connected in series with the five 2-fold up-sampling modules in the decoder.
Dependencies are inferred from the claim text.