Capacity analysis of voting networks with application to human face recognition

Human face recognition is growing field of biometrics in which biological features are used to identify, individuals. In face recognition, the human face is the biological feature used to perform identification. This field has recently received an increased amount of attention because of potential applications such as security systems, law enforcement agencies, ATM/credit card applications, and human computer interfaces.; The thesis is divided into two sections. In the first part, we study the capacity analysis of binary images using 2-level decoupled Hamming network which is a high performance discrete time/discrete state associative memory model. This model is a generalization of Hamming memory in the sense that it provides the local distance computations in the first level and a voting mechanism in the second level. We will study the effect of system dimension, window size, and noise on the capacity and error correction capability of the two-level decoupled Hamming memory.; In the second part, we develop a fully automated face recognition system and study the classification, false positive, and temporal performance of the system using a database consisting of face images of 1200 individuals. Two techniques are developed for this purpose. The first one is a recognition algorithm based on the wavelet transform. The second one is a generalization of the 2-level decoupled Hamming network in which the local windows are optimally repositioned on a face in order to compensate for facial expressions, and small degrees of translation, tilt and rotation which is called voting with dynamic local templates. This 2-level network with its dynamic templates nature is able to simultaneously achieve a high correct classification and low false positive rate. The well known nearest neighbor algorithm is used to benchmark our results.; The proposed voting with dynamic local templates gave 2.5 percent rejection with 0 percent misclassification on database of 1000 individuals.