Combine the clustering algorithm and representation-based algorithm for concurrent classification of test samples

Sparse representation (SR) is a novel pattern recognition method. The algorithm of SR usually performs well. However, in processing a massive concurrent recognition task, SR has a very high computational cost because every test sample has to seek to an optimal linear combination of all the training samples. To this end, we propose a novel method which can perform well without needing to seek a linear combination of all the training samples for every test sample. Our proposed method can be divided into two steps: the first step of the proposed method uses c-means clustering to categorize the test sets into c subsets and then calculates K nearest neighbors for each class centre from all the training samples. The second step represents test samples located in each subset as a linear combination of the according K nearest neighbors and uses representation result to perform ultimate classification. A large number of experimental results show that the proposed algorithm is promising.