| Summary: | © 2017, Springer Nature Singapore Pte Ltd. The K-SVD algorithm aims to find an adaptive dictionary for a set of signals by using the sparse representation optimization and constrained singular value decomposition. In this paper, firstly, the original K-SVD algorithm, as well as some sparse representation algorithms including l 0 -norm OMP and l 1 -norm Lasso were reviewed. Secondly, the revised Lasso algorithm was embedded into the K-SVD process and a new different K-SVD algorithms with l 1 -norm Lasso embedded in (RL-K-SVD algrithm) was established. Finally, extensive experiments had been completed on necessary parameters determination, further on the performance compare of recovery error and recognition for the original K-SVD and RL-K-SVD algorithms. The results indicate that within a certain scope of parameter settings, the RL-K-SVD algorithm performs better on image recognition than K-SVD; the time cost for training sample number is lower for RL-K-SVD in case that the sample number is increased to a certain extend.
|
|---|