Complex wavelet transform with vocabulary tree for content based image retrieval

Abstract

In this paper, combinations of spatial orientation tree (SOT), two-dimensional complex wavelet transform (CWT) and vocabulary tree (VT) is used for feature collection and retrieval of the images from natural as well as texture image database. SOT represents the parent-offspring relationship among the wavelet coefficients in multi-resolution wavelet sub-bands. Similarly, CWT captures directional information more accurately as compared to discrete wavelet transforms (DWT). SOT gives the set of descriptor vectors for each image which are further indexed by using vocabulary tree. The proposed method is tested on Corel 1000 and texture image database (Brodatz and USC) and the retrieval results have demonstrated a significant improvement in average precision, average recall and average rank compared to complex wavelet transform (CWT), optimal quantized wavelet correlogram (OQWC), Gabor wavelet correlogram (GWC).