Robust target localization and segmentation using statistical methods

This thesis aims to contribute to the area of visual tracking, which is the process of identifying an object of interest through a sequence of successive images. Some of the challenges associated with these tasks are image noise, occlusions, background clutter, complex object shapes, etc.The work contained in this thesis explores kernel-based statistical methods. These methods map the data to a higher dimensional space where the tasks of classification and clustering are easily carried out. There are two problems related to the mapping: The out-of-sample and the pre-image problem. A pre-image framework for some of the manifold learning and dimensional reduction methods is developed.Two algorithms are developed for visual tracking that are robust to noise and occlusions. In the first algorithm (Chapter 3), a KPCA-based eigenspace representation is used. The de-noising and clustering capabilities of the KPCA procedure lead to a robust algorithm. This framework is further extended in Chapter 6 to incorporate the background information in an energy based formulation, which is minimized using graph cut. Chapter 7 extends this framework to track multiple objects using a single learned model.In the second method, a robust density comparison framework is developed (Chapter 5) that is applied to visual tracking (Chapter 8), where an object is tracked by minimizing the distance between a model distribution and given candidate distributions.The superior performance of kernel-based algorithms comes at a price of increased storage and computational requirements. A novel method is proposed in Chapter 4, that takes advantage of the universal approximation capabilities of generalized radial basis function neural networks to reduce the computational and storage requirements for kernel-based methods. The ideas developed are general and are applicable to other kernel-based methods, such as KPCA and support vector machines.