P‐8.6: A Progressive Single Image Super‐Resolution Algorithm

The existing single‐image super‐resolution convolutional neural network usually elevates the resolution to the specified scale in one step in the final image reconstruction part. However, this may lead to the incomplete reconstruction of some detailed features. Therefore, we propose a super‐resolution reconstruction algorithm based on progressive architecture. For large‐scale scaling, high‐quality images can be gradually generated from coarse to fine in progressive reconstruction. An adaptive high‐frequency residual module is designed to learn the image features through two branches of low frequency and high frequency, respectively, and the output is adaptive fusion. This process is embedded in a Features cut Block to refine feature extraction during reconstruction, capture the dependency between the LR‐HR image pairs, and introduce an efficient channel and spatial attention mechanism to fuse the image information. Experimental results show that this algorithm has good performance and can enhance the reconstruction ability of images.