Rotation-invariant texture retrieval using wavelet-based hidden Markov trees

In this paper, we present a novel approach for rotation-invariant texture retrieval using multistated wavelet-based hidden Markov trees (MWHMT). We propose a new model to capture statistical dependencies across three independent wavelet subbands. The proposed approach has been applied to CBIR application, rotation-invariant texture retrieval. The feature extraction of the texture is then performed using the signature of the texture, which is generated from the wavelet coefficients of each subband across each scale. We used Kullback–Leibler (KL) distance measure to find the similarity between textures. We have tested our approach for Brodatz texture database and evaluate the retrieval performance in terms of precision and recall. The experimental results show that the proposed method outperforms earlier wavelet-based methods.