HFFR-SR: Hierarchical Fusion Feature Representations for Super Resolution of Old Images
Yiwei Jia, Shuo-Yu Huang, Xueming Li, Xianlin Zhang · 2023
The resolution of traditional old images can not meet the needs of modern playback devices as the long history and the low-level equipments. Therefore, super-resolution reconstruction for old images is necessary. In this paper, we propose an HFFR-SR net which integrate the hierachical feature representations for addressing this issue. Specifically, (1) a novel transformer basic block is designed which can improve the scope of receptive field; (2) the lightweight attention module CBAM is constructed for further restoring the high-frequency details of old images. The experimental results on large-scale datasets show that the proposed HFFR-SR model can achieve satisfactory results in both quantitative and subjective quality comparison.