A METHOD OF CLASSIFICATION FOR K-NN BASED IMAGE RETRIEVAL

Abstract

In this paper, a stratified data approach was applied to a similar image search problem through a bag vision feature from BoVW (Bag of Visual Words). The classification method is based on the k-NN (k-Nearest Neighbor) algorithm with the input data being a feature vector
of the image. From an initial image data set, we construct a bag of visual words to stores images that are substantially similar in content. After classifying the input image by the k-NN method, a set of similar images is extracted from BoVW. In the k-NN method, in addition to k nearest neighbors, a radius θ is used to statistically classify the image. Each BoVW links to other word bags through its representative semantic class. Experiments were built on COREL image database (1,000 images) to evaluate the accuracy and compare with other related works on the same data set. According to empirical results, our recommendations are effective and can be applied in various multimedia systems.