Stacked Auto-Encoders Based Biometrics Recognition

Abstract

Recently deep learning has shown significant achievement in the performance of many tasks, like natural language processing, image and speech recognition. Also, this improvement concerns multiple biometrics recognition systems. In this work, we focus on biometrics recognition, we present a stacked auto-encoder-based approach for various biometrics recognition, including Iris, Ear, palm-print, and face recognition. The proposed method allows training a neural network that includes two hidden layers for biometrics tasks. It runs in two steps, in the first one, each layer is trained individually in an unsupervised manner by auto-encoders, then the layers are stacked and trained in a supervised way. Experimental results on images, obtained from publicly available biometrics databases clearly demonstrate the benefit of using stacked auto-encoders as feature extraction and dimension reduction tools for biometrics recognition, as significant high accuracy rates are obtained over the four databases.