Content frequency and shape features based on CBIR: application to colour images

Abstract

Due to the diversity of the image content, we propose in this study a new technique for content-based image retrieval to characterise the image. In this scenario, all images are characterised by their frequency content and their shape information. Indeed, using the high-resolution spectral analysis methods, especially the 2-D estimation of signal parameters via rotational invariance techniques, we extract from the image its content frequency and with the angular radial transform its shape information to construct a new vector descriptor. The experimental results applied to the Coil_100 database show the effectiveness and the robustness of our approach compared with the method based on the shame information only, and the precision average can be reached as 74.49%.