Tree-based multiscale pursuit

This paper introduces a new practical algorithm called tree-based multiscale pursuit (TMSP) for sparse signal representation over redundant dictionary. The main contribution is the construction of the tree-structured multiscale dictionary which is based on the similarity of the atoms. The similarity is estimated by the distance of the atoms´ parameters. Owing to efficiently organizing the original dictionary, TMSP algorithm makes a good compromise between the high performance and efficient implementation, and can satisfy the users´ different performances requirements by simply stopping the pursuit at different levels of the tree. Moreover, the tree-structured multiscale dictionary requires no extra memory and is easier to be constructed when compared with the tree-structured dictionary used in the tree-based pursuit.