Multi-stream Convolutional Neural Networks Fusion for Palmprint Recognition

Abstract

In recent years, researchers have carried out palmprint recognition study based on deep learning, and proposed a variety of methods based on deep learning. In these methods, convolution neural networks (CNN) were directly applied to the original ROI image of palmprint for training and recognition. In fact, after processing, palmprint can have other representations, such as directional representation, and magnitude representation, etc. However, researchers have not investigated the problem that applied CNNs to other representations of palmprint for recognition. In this paper, we propose a novel framework of multi-stream CNNs fusion for palmprint recognition. In this framework, palmprint are firstly processed into other different representations. Next, CNNs are applied to different palmprint representations for recognition, and then, the information fusion is conducted to effectively improve the recognition accuracy. Under this framework, we propose a concrete implementation, i.e., three-stream CNNs fusion for palmprint recognition. We evaluate the proposed method on five palmprint database. Experimental results show that the recognition accuracy of the proposed method is obviously better than some classical traditional methods and deep learning methods.