Algoritmy pro klasifikaci visuálních signálů s využitím technik extrakce významných rysů

Abstract

Classification of visual data is a fundamental task in state-of-the-art computer vision. The efficiency of the commonly used neural networks crucially depends on a large amount of training data. This thesis explores alternative conventional classification methods, where the dimension of the visual data is reduced by projecting the data onto a space of significant features. The local and global feature extraction techniques, namely SIFT, SURF, ORB, and PCA are compared. The classification by SVM and the Bayesian Model is examined on three datasets with a small number of training images of different complexity.