Efficient Texture Image Retrieval using Prims Algorithm with K-Degree Nearest Neighbor Graph

Abstract

In this paper, we have tried to improve texture image retrieval performance by introducing post processing based on the greedy technique called Prims algorithm with K-degree Nearest-Neighbor graph (K-DNN). The proposed method uses the sparse representation of K-DNN graph called distance matrix instead of feature database directly, which is the distance between every K nearest images to the every other K images of the database. Due to K-degree Nearest-Neighbor sparse graph representation, the retrieval performance of the proposed method is superior over the traditional CBIR. The proposed system is tested with two different image descriptors. First image descriptors are formed with combination of rotated complex wavelet filters (RCWF) and dual tree complex wavelets (DT-CWT). Second image descriptors are formed with Contourlet Transform (CT).