Extended ERDBS: Efficient Search for Image Super-Resolution Using Residual Dense Blocks

The remarkable progress in deep convolutional neu-ral networks has significantly advanced the field of single-image super-resolution. However, the practical implementation of deep learning approaches faces challenges related to computational and memory requirements, especially on mobile devices. To over-come these challenges and address the need for rapid, efficient, and precise networks designed for enhancing image resolution, this paper presents an Extended and ERDBS algorithm. This algorithm aims to improve the performance of image super-resolution by leveraging guided evolution techniques. Addition-ally, we have incorporated some changes to the mutation and training processes during the evolution stage. The experimental results confirms the effectiveness of these modifications, with an average PSNR of 41, and as they enhance the overall performance of the proposed approach also approach to a better super-resolution image.