Neural network-based teeth recognition system using hybrid features

Abstract

Nowadays, biometric technology is used in various security applications. The efficiency of such applications depends upon a type of biometric information. Nevertheless, some information can be faked by intent surgery or they are unexpectedly reshaped such as face, iris, palmprint and fingerprint. Unlike ordinary features, teeth cannot be easily reshaped. In this thesis, hybrid features and machine learning model for teeth recognition are proposed. Hybrid features of this system are composed of global and local features simultaneously fed into the system. In this thesis, proposed global features composed of singular values from singular value decomposition and color histogram of teeth image are analyzed and give the adequate result whilst the proposed local features are the ratio of the width from upper-front-teeth. These features were fed into the multilayer perceptron network with Levenberg-Marquart backpropagation training algorithm. With these features and model, the proposed method performs better than other existing techniques in terms of accuracy and error.