Super-resolution using POCS-based reconstruction with artifact reduction constraints
Jun-Yong Kim, Rae‐Hong Park, Seungjoon Yang · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2005
In this paper, we propose a super-resolution (SR) algorithm for reconstructing a high-resolution (HR) image from multiple low-resolution (LR) sequence images, in which the projection onto convex sets (POCS) algorithm is employed with artifact reduction constraints. In real dynamic sequences having the object-motion occlusion or relatively large motions, artifact reduction as well as enhancement of image quality is needed. The proposed POCS-based algorithm reduces the degradation caused by the motion compensation (MC) error, where the motion confidence map is used to find the LR pixel with a relatively small MC error. The decision bound in the motion confidence map corresponds to the bound in residual computation in the conventional POCS algorithm. Also, the proposed algorithm utilizes the directional information at edges in reconstructing the SR image, in which the four-directional projection function is used to resolve the desired SR pixel. Finally, the over-compensation is avoided in the iteration process by adding the appropriate constraint. Experimental results with several test sequences show the effectiveness of the proposed algorithm.