An Analysis on the Two-phase Test Sample Sparse Representation Method and an Improved Method

摘要

The recently proposed two-phase test sample sparse representation (TPTSR) method makes a great contribution to the field of face recognition. Though TPTSR uses a computationally very efficient algorithm, it can obtain a better performance than the well-known sparse representation method. In the first phase of TPTSR, the determined M nearest neighbors for the test sample seem not to be optimal in terms of the representation error. In other words, it is probably that there exist M training samples whose linear combination has a smaller deviation from the test sample. This deviation is referred to as representation error of the test sample. If the smaller the representation error, the higher the accuracy, then it will be possible that one can obtain a better face recognition result. In this paper, in order to explore this issue, we propose an improved method. This method revises the first phase of TPTSR as a step that uses the global search algorithm to determine the M " optimal " nearest neighbors of the test sample. We show the representation error and classification accuracy of the improved method and TPTSR by experimental results.