Comparative Effect of Distance Metrics on Selected Texture Features for Content-Based Image Retrieval System

Abstract

The effect of distance metrics on the retrieval performance of CBIR system characterized by texture features was evaluated in this paper. Tamura and Gabor Wavelet transform texture features were used to create the feature vector database and to match 4 sample query images respectively with the image database of 400 and 600 images of four different classes Euclidean distance, Manhattan distance, Cosine angle distance, Quadratic-form distance and Pearson correlation were used. The retrieval performance was evaluated using the average precision with the distance measures. Evaluated results from the CBIR system algorithm shows that Manhattan distance metric gave the best retrieval performance on the used two set of image database.

Keywords: Content-Based Image Retrieval, Feature extraction, Distance metrics