image classification using BoWs

What is it about?

While extracting the information of digital images is very simple and intuitive for human being, it is very complicated for computer systems. As a result, any designed algorithm encounters many challenges to explore the images content. We automatically classified the content of images to a number of predefined concepts. Our method works well and competitive to the results of others in different image collections like natural scenes and pictures of single object.

Featured Image

Why is it important?

We utilized spatial information to improve image classification accuracy. At the same time we kept the time and space complexity of algorithm low by using a hierarchical tree structure defined on visual words of the dictionary. That is, instead of using all visual words, bi-gram terms are constructed based on informative internal nodes of this tree. Our findings show improvement in accuracy by adding just few bi-gram terms to the BoWs histogram.