Abstract: Rececnt researches on single-image super-resolution based on deep convolutional neural networks have achieved remarkable results. However， with increasing neural network scale， to reduce the network constructing difficulty and the computation costs has become a difficult issue. To solve this problem， a dual-channel multi-perception convolutional neural network (DMCN) is proposed. Specifically， the dense skip connection between two convolutional channels with different sizes of convolution kernels is established. Then， an adjustable inter-layer fusion structure is established between different layers. This structure makes it possible to learn abundant image feature information through a small-scale convolutional neural network. The experimental results show that the proposed model outperforms most of the typical algorithms.

Key words: single-image super-resolution, dual-channel multi-perception convolutional neural network(DMCN), dense skip connection, residual network, deep learning