Neurobiologically-based System for Object Segmentation and Recognition

This phase I project will create, for the first time, a dynamic system, based on neurophysiology of the brain, to perfrom image segmentation using high level info. In computer vision, pattern recognition is frequently hampered by poor image segmentation. Segmentation algorithms have not been able to segment an image as a human would, instead they form regions based on mathematically defined criteria. Because existing segmentation algorithms are underdeveloped, the segments they create do not benefit pattern recognition systems, which have enough problems recognizing an object, let alone recognizing an object that has not segmented accurately. In IOS' dynamic system, low level sgementation results will feed forward omto an associative memory. the associative memeory will feedback to the first level to adjust the segmentation results. Merging both tasks in a dynamically interacting system will create a powerful and accurate image segmentation tool for object recognition and machine vision applications.