Fast Full Search Equivalent Block Matching for Multichannel Images


Block matching is a fundamental tool to search blocks (patches) similar or identical to a given query in image processing. Generally, a full search (FS) algorithm is the most accurate but requires vast computation especially in multichannel images, where the data volume is increasing due to higher definition and more channels. In this paper, we present a fast FS-equivalent algorithm using orthonormal tree-structured Haar transform (OTSHT) for multichannel images. We demonstrate the superior performance of three-dimensional OTSHT comparing with state-of-the-art algorithms. This significant speedup can enable new applications of block matching in multichannel images.

IEICE Information and Communication Technology Forum