2022 年 10 巻 3 号 p. 130-139
Various techniques have been proposed for restoring JPEG compressed images degraded with inevitable artifacts. Convex optimization is often used for such techniques, where a prescribed function is minimized over the narrow quantization constraint set (NQCS) characterized by a scaling parameter. Although the scaling parameter significantly affects the restoration quality, the optimal determination of the scaling parameter is actually difficult because it is highly dependent on the target JPEG compressed image. To overcome this difficulty, we propose a novel NQCS-based artifact reduction framework requiring no explicit parameter optimization. In our framework, a set of candidate images is first obtained, whereupon the restored image is computed by applying simple statistical operations for the set. The advantages of our approach are demonstrated by comparing it with several existing approaches in terms of the theoretical error bounds and experimental restoration results.
