Content-based Image Retrieval Using Moment-preserving Edge Detection

Abstract

A content-based image retrieval algorithm based on a new edge detection technique is proposed. Both the query and database images are divided into non-overlapping square blocks and coded by the mean in each uniform block and by edge information in each non-uniform block. The coded blocks of a query image are then used to find matches from an image database. The edge feature in a given block is detected by applying the moment-preserving principle to the image data. The edge directions are approximated by multiples of 45º to speed up the matching process without introducing obvious distortion. For a larger database, a selective filtering strategy based on the visual-pattern histograms is also described to further speed up the retrieval process. The solution to the edge detection problem in a given block is also analytic. This algorithm can be performed very fast for large-database applications with no need for special hardware.