Description

ABSTRACT:

                We have provided a comprehensive comparison of several popular 1-min algorithms in sparse representation based classification. We have shown that the ALM algorithms compare favorably to other classical and accelerated sparse optimization methods, especially when applied to recognizing high-resolution face images. In particular, the dual ALM algorithm performs the best in the face recognition experiment, and scales well in terms of the number of subjects. Hence it is well suited for large-scale classification problems. The primal ALM algorithm is the fastest method in solving the face alignment problem.