A fast compressive sensing based N-best class selector for classification applications

In this paper, an N-best class selector based on compressive sensing (CS) algorithm and a tree search strategy is introduced and applied for classification applications and its accuracy and complexity are compared with some well-known classifiers. In this approach, classification is done in three steps. At first, the set of most similar training samples for the specific test sample is selected by KD-tree search algorithm. Then, a CS based N-best class selector is used to limit the classifier input into certain classes. This makes the classifier adapt to each test sample and reduces the empirical risk. Finally, a well known low error rate classifier is used to classify the candidate classes. By this approach, we obtain competitive results with promising computational complexity in comparison with state of the art classifiers which causes this approach become a suitable candidate in common classification problems.