Deep belief networks for iris recognition based on contour detection

Iris recognition has proved to be one of the most reliable and stable biometric for human identification. This paper outlines an iris recognition approach based on deep learning. In addition, contour based feature vector has been used to discriminate samples belonging to different classes i.e. difference of sclera-iris and iris-pupil contours, and is named as “Unique Signature”. Moreover, contours of both the boundaries are extracted using the radius vector function. The process of feature extraction starts with preprocessing i.e. removal of specular highlight, then weighted centroid of the eye is used as the reference to extract the contour points. Further, the aforementioned approach is rotation, translation and scale invariant. Also, deep belief network (DBN) with modified back-propagation algorithm based feed-forward neural network (RVLR-NN) has been used for classification. Simulation results have been observed for the benchmark CASIA iris database i.e. CASIA-Iris-V4. Further, classification accuracies have been observed at different levels of signal-to-noise ratio (SNR) i.e. additive white Gaussian noise (AWGN). Also, comparisons have been made with other techniques using contour based features for iris detection. Further, all simulations have been performed in MATLAB. The proposed approach outperforms both the neural network based single and dual boundary approaches in terms of classification accuracies. Observed results have shown good classification accuracies especially at high noise levels.