Comparison of some Bag-of-Words models for image recognition

Abstract

There are a number of methods for image recognition and they are mainly based on Bag-of-Words (BoW) model. These models can be divided into two categories: generative and discriminative models. Some of the generative models are Naïve Bayes, latent Dirichlet allocation and Probabilistic Latent Semantic Analysis. Discriminative methods are Nearest neighbor classification, Support Vector Machines and Pyramid match kernel. Goal of this paper is to compare two implementations of Support vector machines model: linear SVM and Hellinger classifier. These two models are compared in simulated environment. Comparison is made by analysing accuracy, speed and processor power consumption measured in simulation.