Image super-resolution with multi-channel convolutional neural networks

This paper proposes image super-resolution techniques with multi-channel convolutional neural networks (CNN). In the proposed method, output pixels are classified into four groups depending on their positions. Those groups are generated from separate channels of the CNN. Finally, they are synthesized into a 2-2 magnified image. This architecture can enlarge images directly without bicubic interpolation. Experimental results have shown that the average PSNR for the proposed method achieves 36.88 dB, which is 0.39 dB higher than that for the conventional SRCNN.