Interscale statistical models for wavelet-based image retrieval

In this paper, we are interested in image indexing in the wavelet transform domain. More precisely, the salient features of the image content correspond to the parameters of the statistical distribution model of the wavelet coefficients. The contribution of our work is twofold. Firstly, a very versatile multivariate interscale distribution driven by the copula theory is chosen to model the joint distribution of the homologous wavelet coefficients considered at different scales. Secondly, the search procedure associated with any request is accelerated through a tree structured search in the features space. Experimental results show that considering interscale information drastically improves the search performances.