Content Based Image Retrieval by Combining Features and Query-By-Sketch

This paper reports an approach to improve content-based image retrieval systems. Most current systems are based on a single technique for feature extraction and similarity search. Each technique has its advantages and drawbacks concerning the result quality. Usually they cover one or two certain features of the image, e.g. histograms or shape information. To overcome these restrictions a flexible framework is proposed, capable of combining several different features in a single retrieval system. This system allows an administrator to build a repository managing different feature vectors. A user searching through this repository defines and weights these features according to his needs in the query. It concludes that a combined retrieval can be used much more widely than a highly specialized one and the use of query-by-sketch or -example combined with semantic information (e.g. keywords) could enhance the result quality.