Local features and high-frequency information double-enhanced network for CT superresolution reconstruction

Le résumé fourni par la source

Aiming at the problems of poor recovery of high frequency information and insufficient capture of local features in transformer-based image super resolution reconstruction, a CT image super resolution reconstruction network with double enhancement of local features and high frequency information is proposed. Firstly, the enhanced channel attention block is embedded in the local feature extraction. In this process, the features in different directions in the shift convolution are fused with the channel features through the attention diagram to expand the scope of local feature extraction and improve the ability of local feature extraction. Secondly, the dual frequency channel self-attention block is designed and introduced into the multi-scale self-attention block to strengthen the feature extraction and feature representation of high-frequency information. The module fuses the spatial dimension features with the attention features in the high-frequency path, and then splicing the features of the low-frequency path to achieve deeper feature fusion and feature extraction. Experimental results show that the proposed method has the best peak signal-to-noise ratio and structural similarity index compared with the current mainstream super-resolution reconstruction methods, and the visual effect is also improved.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.