Recently, Associate Professor Chen Liang and her team from college of Photonic and Electronic Engineering of our school got four papers published on the theoretical research on face super-resolution (FSR) related to the generalization ability of heterogeneous images in the multimedia flagship journals IEEE Transactions on Image Processing（CCF A）and IEEE Transactions on Circuits and Systems for Video Technology（CCF B）recommended by China Computer Federation. In light of the complex degradation process of extremely low-quality input images under actual conditions, the team carried out the research from the two perspectives of the robustness enhancement of the algorithm itself and the removal of heterogeneous degradation factors, and built a fast processing mechanism based on small samples in terms of the supervision of context information, the introduction of a multi-layer nearest neighbor network, and the construction of a new framework for degradation to improve the robustness of the algorithm.
Among them, in the first two research results the team propose assisting the reconstruction of the target face image patches with the context structure distribution of the face as supervision and mitigating the effects of the heterogeneous degradation on the local image patches by expanding the receptive field from the local to the global face to improve the reconstruction accuracy of the local patches. The two research results were published in the journal IEEE Transactions on Image Processing (CCF A), titled Robust Face Super-Resolution via Position Relation Model Based on Global Face Context and Robust Face Image Super-Resolution via Joint Learning of Subdivided Contextual Model respectively, with Fujian Normal University as the first unit of the papers, and Associate Professor Chen Liang as the first author.
The first paper link: https://ieeexplore.ieee.org/document/9199538
The second paper link: https://ieeexplore.ieee.org/document/8733990
In the third research result the team propose introducing a multi-layer nearest neighbor network to provide highly correlated neighbor candidate sets for the reconstruction process and enhance the representation ability of heterogeneous image patches, and mitigating the effects of the heterogeneous degradation on the local image patches by expanding the receptive field from the local to the global face for better reconstruction results. The work was published in the journal IEEE Transactions on Circuits and Systems for Video Technology (CCF B), titled Modeling and Optimizing of the Multi-layer Nearest Neighbor Network for Face Image Super-Resolution, with Fujian Normal University as the first unit of the paper, and Associate Professor Chen Liang as the first author.
The third paper link: https://ieeexplore.ieee.org/document/8717720
In the fourth research result the team propose a reconsideration of providing sub-patch sets with high coupling for the reconstruction process in light of the framework for the complex heterogeneous degradation process, and by removing the heterogeneous degradation factors, in order to enhance the representation ability of heterogeneous image patches, and assist the reconstruction of target face image patches for better reconstruction results. The work was published in the journal IEEE Transactions on Image Processing (CCF A), titled Multi-stage Degradation Homogenization for Super-Resolution of Face Images with Extreme Degradations, with Fujian Normal University as the first unit of the paper, and Associate Professor Chen Liang as the first author.
The fourth paper link: https://ieeexplore.ieee.org/document/9451563
The above work was funded by the National Natural Science Foundation of China, the Joint Fund Project to Promote Cross-Straits Science and Technology Cooperation of National Natural Science Foundation of China, and the Natural Science Foundation of Fujian Province and so on.
(Translated by Zheng Siming/ Reviewed by Xie Xiujuan)