Sparse Proximity based Robust Fingerprint Recognition

The primary objective of this paper is to explore the applicability of sparse representation based classification (SRC), particularly at the fingerprint recognition problem. This paper proposes sparse proximity based fingerprint matching methodology. The sparse representation based classification problem can be solved as representing the test sample in terms of training set with some sparse residual error which is solved through sparse coding problem. The sparse representation based classification methodology learns an over-complete dictionary using the discriminating features extracted from the training samples. The learned over-complete dictionary contains each column as the feature vector of one of the training samples. The key concept behind SRC is that if sufficient training samples are available for each class in training phase then it will be possible to represent test samples as a linear combination of training samples belonging to that particular class. Sparse proximity error is calculated for a test sample corresponding to dictionary of each class. The test sample is labeled with the class whose dictionary yields minimum reconstruction error. The experimental results show that the proposed method achieves higher recognition accuracy in comparison with other state of the art methods.