An analysis of BoVW and cBoVW based image retrieval

Abstract

Bag-of-Visual Word (BoVW) is one of the popularly used Image Representation model. It is a sparse vector model based on the occurrence count of the visual words extracted from image features. One of the major limitations of BoVW model is the quantization error that is caused mainly because of false matches. In this paper, first the BoVW model based Image Retrieval is performed experimented and analyzed for various visual word size and vocabulary sizes. The novelty of this paper is based on the assumption that objects of our interest covers mostly the centre part of the images. In this paper, an algorithm where features belonging to the centre part of the images are given attention (cBoVW) and applied for online query Image Retrieval. Using CalTech 256 dataset, several evaluations are done and results seem to be promising in resolving quantization error when compared with the conventional BoVW model.