15th European Conference on Artificial Intelligence
|July 21-26 2002 Lyon France|
Marco Anelli, Alessandro Micarelli, Enver Sangineto
We propose a new version of the famous Ballard's Generalized Hough Transform (GHT) for image retrieval by shape similarity. Indeed, the GHT is a very powerful pattern recognition technique, robust to noise and occlusion situations, utilized in hundreds of different machine vision problems. Nevertheless, it is conceived for an exact matching between the model and the input image, while, in image retrieval, the user description of a figure is inherently abstract and approximate, thus locally different from each data base image. In this paper we present a version of the GHT locally tolerant to deformations which successfully fits image retrieval peculiarities without accuracy loss. The proposed method has been implemented and tested using images randomly chosen from the Web with very good experimental results.
Keywords: Vision, Information Retrieval, Information Extraction
Citation: Marco Anelli, Alessandro Micarelli, Enver Sangineto: A Deformation Tolerant Version of the Generalized Hough Transform for Image Retrieval. In F. van Harmelen (ed.): ECAI2002, Proceedings of the 15th European Conference on Artificial Intelligence, IOS Press, Amsterdam, 2002, pp.721-725.