A Content Based Image Retrieval System Using K-means and K-means++

Abstract

In recent years, there is an increasing research focus on Bag-of-Words based visual search. The state-ofthe- art in visual search systems are built based upon a visual vocabulary model with an inverted indexing structure. This search paradigm is achieved by systems that are inspired by text retrieval. Visual words are image descriptors selected from a discrete vocabulary. These words are mapped from highdimensional descriptors characterizing local regions of images. In this paper, we address problems of visual vocabulary constructing by comparing k-means with kmeans++ algorithms. As expected, k-means++ is clearly superior to the classical algorithm k-means both in terms of quality and running time