Actually, there is several models of recovery of information in audio, video, image that are based in characteristic such as the color, texture, shapes and relationships space. Although most of these works are guided to the vectorial extraction of characteristic and the use of techniques of classification as neural nets nets and data mining, only a limited attention was given to the combination of techniques of vectorial characterization and models of similarity indexed in metric spaces. In this paper, we present the recovery for similarity of shape by using local characteristics and indexation in metric structures of fingerprints. The shape is processed in dimensions that are related with the directions, positions and radial neighbourhood of characteristic points. The indexation of those fingerprints is implemented through the insertion of vectors in the MTree. To solve problems related to the low quality of the image we uses the Fourier transform and Inverse Fourier transform with Gaussian low pass filters Our experiments suggest that the use of range queries on metric spaces solve the problems of high consumption of CPU time of the neural nets and lingering times of queries associated to the neural nets and data mining algorithms.
|