Efficient Shape and Color Based Methods and Architecture for Content-Based Visual Retrieval

An alternative to text-annotated image databases is to use descriptors based on inherent Image properties. Indexing and retrieval baaed on visual content overcome the deficiencies of text annotation {e.g., database size, automatic feature indexing, language mismatch issues) and facilitate automatic handling of data [1]. Semantic-preserving image compression techniques result in descriptor coefficients representing Image attributes, without the need for a domain expert. Automatic feature extraction is computationally expensive if dynamic operation is required. This paper presents new methods to address those problems, including new feature encoding algorithms for shape indexing, new analysis tools for color data manipulation, and Improved shape retrieval techniques for efficient database searching. A new object-based hybrid multiprocessor architecture Is used to design a processor for content-based video indexing.