Robust texture image retrieval using hierarchical correlations of wavelet coefficients

We propose a robust texture image retrieval using hierarchical relations between the decomposed sub-images based on the wavelet transform. The key idea is to use hierarchical correlations between the wavelet coefficients as texture features. We express the pyramidal structure of wavelet coefficients by associating the nodes of a complete quadtree with wavelet coefficients. We define a hierarchical dissimilarity vector between a parent node and his child, to express a hierarchical relation between them. Then, to describe a relation among child nodes, we compute a covariance matrix of dissimilarity vectors. We associate the covariance matrix with the parent node, and call such a quadtree “hierarchical correlation wavelet tree” (HCWT). Finally, to make up an index of database, we calculate the texture feature vectors defined by the diagonal of elements of the covariance matrix of HCWT and the texture feature vectors, and use the discriminant analysis to make an effective index from the texture feature vectors. For retrieving the similar images, we use the k-nearest neighbor search in the index space. Experiments showed a good performance of the proposed retrieval method