A Self-ensemble Approach for Noise and Compression Artifacts Removal using Convolutional Neural Network
Byeongyong Ahn, Gu Yong Park, Yoonsik Kim, Nam Ik Cho · IEIE Transactions on Smart Processing and Computing · 2018
There have been many discriminative learning methods using convolutional neural networks (CNN) for image restoration problems, which learn the mapping function from a degraded input to the clean output. In this paper, we propose a self-ensemble method that can find enhanced restoration results from the multiple trials of a trained CNN with different but related inputs. Specifically, it is noted that the CNN sometimes finds different mapping functions when the input is transformed by a reversible transform and thus produces different but related outputs with the original. Hence averaging the outputs for several different transformed inputs can enhance the results as evidenced by the network ensemble methods. Unlike the conventional ensemble approaches that require several networks, the proposed method needs only a single network. Experimental results show that adding an additional transform usually brings additional gain on image denoising and artifacts removal problems.