An Improved Two-Phase Sparse Representation Method for Traffic Sign Recognition

Abstract

The two-phase test sample sparse representation (TPTSR) is a method that performs very well on face recognition. But with the increasing of training samples, it may cause memory overflow when we do matrix operations. To solve this problem, we propose an improved TPTSR method which is based on local dictionary. In the first stage of TPTSR, we split the redundant dictionary that is combined by all training samples into local dictionaries and solve the linear combination of each local dictionary for test sample, then select M nearest neighbors of the test sample from local dictionaries. In the second stage, we represent the test sample as a linear combination of M nearest neighbors and use the representation result to do classification. The experimental results show that this algorithm is superior to the traditional algorithms such as PCA, LDA and OMP. Its recognition rate can reach 94.2%.