Region-based image retrieval using a joint scalable Bayesian segmentation and feature extraction
Sarra Sakji-Nsibi, Amel Benazza‐Benyahia · 2016
In this paper, a region based system is designed for textured image retrieval. A scalable joint Bayesian segmentation and feature extraction in the wavelet transform domain is performed. The segmentation map and the extracted region features are refined by exploiting more decomposition levels. In order to account for spatial dependencies, Markov Random Field (MRF) is employed to model the prior distribution of the segmentation map at each scale. Moreover, a coarse to fine resolution retrieval procedure is proposed. Experimental results carried out on remote sensing images corroborate the gain achieved by the proposed indexing method. Moreover, the resort to an adaptive smoothing parameter reflecting the image homogeneity improves the gain provided by the proposed approach.