Analysis of descriptors relevant to image search by content

This article investigates the relevance of visual descriptors in content based search. We propose an approach of indexing and searching of images by content based on the extraction of different vectors characteristic of remote sensing images decomposed into wavelets [6]. It is shown that the Coiflets-3 have better performances in terms of results and computation time for images with a strong directional predominance. It is shown that the Daubechies-3 and Symmlet-3 wavelets show better results on asymmetric or random images. The results of this work are compared with previous work and show a clear improvement.