Image Classification using SVM Classifier Learned by AdaBoost Method

Abstract

This thesis presents the algorithm that categorizes images by objects contained in the images. The images are encoded with bag-of-features (BoF) model which represents an image as a collection of unordered features extracted from the local patches. To deal with the classification of multiple object categories, the one-versus-all method is applied for the implementation of multi-class classifier. The object classifiers are built as the number of object categories, and each classifier decides whether an image is included in the object category or not. The object classifier has been developed on the AdaBoost method. The object classifier is given by the weighted sum of 200 support vector machine (SVM) component classifiers. Among multiple object classifiers, the classifier with the highest output function value finally determines the category of the object image. The classification efficiency of the presented algorithm has been illustrated on the images from Caltech-101 dataset.