SPARSE REPRESENTATIONS OF SIGNALS AND THEIR APPLICATION TO PATTERN MATCHING AND INFORMATION HIDING

Abstract

Sparse overcomplete representations of signals have garnered a great deal of attention dur-ing the past decade and a half. The main reasons for this noteworthy interest are their high compression ratios and characteristic energy compaction. These have, however, been primarily been used in the fields of denoising, signal source separation, pattern classifica-tion and information hiding. The present endeavour is devoted to the different aspects of sparse overcomplete represen-tations in the form of a survey of various sparse recovery algorithms and their merits as well as demerits and proposing a sparse recovery algorithm based upon genetic algorithms. This is followed by a discussion of the problem of dictionary learning for sparse represen-tations where a popular algorithm has been examined and modified to obtain improved results. Finally, the applications of sparse representations have been extended to the fields of pattern matching as well as of information hiding and a steganography algorithm has been presented