Texture image retrieval using biorthogonal multiwavelet filters

In this paper, we explore texture image retrieval mechanisms based on wavelet features and propose a novel method of feature extraction based on biorthogonal multiwavelet for texture image retrieval. It can capture texture features more effectively than traditional wavelet. Firstly biorthogonal multiwavelet is designed to decompose texture images. Then features are obtained by computing the energy and standard deviation on each subband of decomposed images. Normalized Euclidean distance metric is used as similarity measure. To check the retrieval performance, texture database of 928 texture images from Brodatz album is created. Experimental results indicate that the proposed method can improve retrieval rate from 64.76% to 72.35% in comparing with uni-wavelet (haar, Db4) and traditional multiwavelet low computation cost yet.