1. ORTHOGONAL MATCHING PURSUIT WITH A NORMALIZED RESIDUAL BASED STOPPING CRITERION
Authors: LING-HUA CHANG , WEN SEN LIU , JIA FU WU
Abstract: Orthogonal matching pursuit (OMP) is a commonly used algorithm in compressed sensing (CS) for estimating a sparse vector/signal x from linear measurements y m , where m n . There are two generally stopping criteria adopted in the iterative OMP. One, assuming the number of nonzero entries of the sparse vector x is known, stop the algorithm after exactly K iterations. The other halt the pursuit if the strength of the residual is smaller than some threshold. These two criteria respectively rely on certain knowledge about the signal and the environment/noise. We propose a normalized residual strength based stopping criterion, which can be employed without the information mentioned above. Numerical results show that under some circumstances, the proposed criterion leads to a smaller normalized signal reconstruction error as compared to that achieved by OMP with exact K iterations and the conventional residual strength based stopping criterion.
Keywords: orthogonal matching prusuit (OMP), compressed sensing (CS), stopping criterion.