SINGLE IMAGE SUPER-RESOLUTION USING ASYNCHRONOUS MULTI-SCALE NETWORK

An existing multi-scale residual network (MSRN) has demonstrated its success in conducting SISR. The MSRN consists of a number of multi-scale residual blocks (MSRBs), and each MSRB performs convolutions with two different sizes of windows. The smaller window extracts features at a low scale, while the larger one achieves that at a high scale. To reduce the number of parameters involved in the MSRB, a new feature extraction module, called the asynchronous multi-scale block (AMB), is proposed in this paper. It is based on the fact that the larger window used in the MSRB can be replaced by two smaller windows without affecting its function. Consequently, by replacing each MSRB with our AMB, an asynchronous multi-scale network (AMNet) is then constructed, which can significantly reduce complexity. This means that more AMBs can be used in our AMNet to deliver superior SISR performance, while maintaining a comparable complexity to the MSRN. To consolidate all image features generated from all scales, a new fusion scheme, called the adaptive feature fusion block (AFFB), is proposed that weights the extracted features according to their importance for further increasing SISR?s performance. Experimental results have shown the superiority of our AMNet when compared with multiple state-of-the-arts.