Bag-of-Features Model, Application to Medical Image Classi?cation

Abstract

The aim of this study is to investigate the Bag-of-Features (BoF) model and its application to medical image classification. With this model, images, or parts of them, are represented as an orderless collection of local image descriptors. A visual codebook is learned from a training set of local image descriptors, usually by performing vector quantization by clustering. Each image is then represented by its distribution of visual words from the codebook. To achieve image classification, a classifier such as a Support Vector Machine (SVM) is used with the obtained image representation as the feature vector. This paper presents a review of the literature on BoF methods and compares the most important implementation choices that have been suggested. In addition, the application of the method to medical image classification is discussed.