Logic models for segmentation and tracking
Mark Christopher Moelich, Tony Fan-Cheong Chan · 2004
The focus of this research is on the development of logic models for segmentation and tracking. Logic models were introduced by Sandberg and Chan, and provide a framework for combining the content of a set of registered images in a single segmentation. The research described in this dissertation improves and extends the logic model framework in several ways. A new set of logic operators are introduced which have good scaling properties, have a well-defined algebra, and converge more quickly than the prototype operators proposed by Sandberg and Chan. A joint segmentation and registration algorithm is developed to relax the restriction that the images be registered. Color logic models are introduced for segmentations based on colors, or features, within a single image. These segmentation models are then used in the development of two tracking algorithms. The first algorithm is effective at tracking a selected object as it moves and deforms within a scene and in the presence of camera motion. The second algorithm assumes a fixed camera, but is able to detect and isolate multiple moving objects in low-resolution video without user input.