A Self-Learning, Non-Geometric, Knowledge-Based Computer Vision System

"Currently knowledge representation models and low-level computer vision algorithms are fundamentally separate domains, where knowledge has no 'meaning' in the visual system itself. Visual recognition algorithms generally attempt to statistically matchpredesigned geometric models or features to images, being limited to narrow applications because of a lack of contextual understanding. The proposed effort aims to implement a revolutionary new method of fundamentally merging knowledge representation withcomputer vision algorithms to produce a computer vision system that is entirely self-learning and functional in any context, without predefined geometric models. It involves a 100% object-oriented model that integrates all three elements of symbolicknowledge representation (frames, a variant of default and modal logics, and forward- and backward-chaining rules) with a one-shot fast-learning approach. This system combines maximum flexibility with efficient and accurate object recognition (includingnon-geometric objects), developing its visual capabilities progressively from low-level to high-level. Proposed work involves the technical design, implementation and testing of this model, and its effectiveness will be demonstrated at the end of Phase Iwith a functioning prototype. Military benefits in the air include vastly improved automatic recognition and comprehension of all types of objects in the air, ground objects, and landscapes. Fast