A machine learning approach to reducing image coding artifacts
Ichiro Matsuda, Tomokazu Ishikawa, Yusuke Kameda, Susumu Itoh · 2017
In this paper, a method for reducing coding artifacts introduced by lossy image compression is proposed. The method is similar to sample adaptive offset (SAO) which is adopted in the H.265/HEVC video coding standard as one of in-loop filtering tools. In the SAO, samples of the reconstructed image are classified into several categories based on some simple algorithms, and an optimum offset value is then added to the samples belonging to each category. Since the classification algorithms are switched on a block-by-block basis, not a negligible amount of side-information must be transmitted to the decoder in addition to the offset values. On the other hand, our method adopts a machine learning technique using a support vector machine (SVM) for the classification process. By applying the common SVM classifier to a whole image, the amount of the side-information can be considerably reduced. Simulation results indicate that the proposed method provides bitrate savings of up to 1.0 % for HD size images degraded through intra frame coding of the H.265/HEVC standard.