Biometric recognition systems employing novel shape-based features

With the increased security requirements in a variety of applications and advances in sensor technology, emerging biometric technologies, including using some lesser known biometrics, have become important research topics. This is due to the potential benefits they may provide as independent biometric markers or as compliments to existing biometric systems. In this work, our aim is to explore new biometric technologies for person identification. We consider three different biometrics, namely, Three Dimensional (3D) and Two Dimensional (2D) ear biometrics, 3D face recognition, and human identification based on dental X-Ray images.;For the ear biometrics component, we propose a novel 3D shape descriptor, termed Histogram of Categorized Shape (HCS), to robustly encode range images within a 3D object detection framework. For the 3D ear detection task, this feature, employed in conjunction with a linear SVM classifier and sliding window technique, produces a robust and efficient 3D ear detection system. Afterwards, we extend the HCS descriptor to an object-centered 3D surface feature descriptor, termed Surface Patch Histogram of Indexed Shape (SPHIS), for local surface patch representation. The SPHIS feature descriptor is evaluated for its effectiveness in real world scenarios where a database may contain ears of highly similar shape. The ear surface is also voxelized to construct a holistic representation. Based on the novel SPHIS feature and the voxelization representation, a unified approach incorporating local and holistic surface features is proposed to improve both the robustness and efficiency of the 3D ear shape matching subsystem, while simultaneously improving the performance of the recognition system. In the 2D domain, a complete, automatic ear biometric system based on 2D images is developed. The color Scale Invariant Feature Transform (SIFT) descriptor is exploited as the feature representation, which in conjunction with a feature fusion method, maximizes the robustness of the recognition system.;For the 3D face recognition component, we propose a method using AdaBoost to determine the geodesic distances between anatomical point pairs that are most discriminative for 3D face recognition. Through a method that establishes a dense set of correspondences between face surfaces, the discriminating potential of geodesic distances between anatomical points is investigated.;For the dental biometrics component, we present a content-based image archiving and retrieval system for assisting in human identification using dental radiographs. The system includes processes for dental image classification, automatic segmentation of bitewing dental X-Ray images, and teeth shape matching.