Garabato: A proposal of a sketch-based Image Retrieval system for the Web

Abstract

A proposal for a queried-by-sketch image retrieval system is introduced as an alternative to text-based image search on the Web. The user will create a sketch as a query that will be matched with the edges extracted from natural images. The main challenge regarding edge detection for Content-based Image Retrieval consists in finding edges for larger regions and avoiding the ones corresponding to textures. For this purpose, a combination of selective smoothing and color segmentation is applied prior edge extraction. An evolutionary algorithm is deployed to optimize the image-processing parameters. Similarity between the user's sketch and the image's edges will be measured regarding two local aspects: spatial proximity and edge orientation. A full architecture for image search on the Web is proposed and preliminary results are reported using a trial database.