Iris Identification Using The Pram Neural Network

Résumé: In this paper we propose a new approach to performer is identification. In contrast to existing approaches that consider the Hamming distance measure to perform identification, the new approach considers the addition of a well trained neural network to identify the iris. The disadvantage of previous schemes is the difficulty to deal with the variability of irises within the same iris class due to noise and movement of the eye as well as difficulties in capturing a clear image of the eye, which makes the choice of threshold values to identify the class to which belong the iris a difficult and a time consuming task. The new approach is based on a digital neural network pRAM. We operate using two alternatives. The first one is based on the application of the raw multi grey level iris image, handled by the bit plane encoding scheme. Whereas, the second alternative is based on the iris code obtained by applying Daugman’s method to represent the distinguishing features of the iris within a binary image. We developed a pRAM net simulator in C++ to handle images taken from a public iris image database. The simulator was exercised with images in an extensive manner. The results obtained are very encouraging as we succeeded to train appropriately the pRAM net and to perform identification with a high identification rate. The obtained results show that identification based on the use of the iris code is more appropriate than applying the normalized multi-grey level iris image.

Publié dans la revue: Courrier du Savoir scientifique et technique