Object Classification Using Machine Learning

Abstract

Computer vision has wealth of research. It spans over image restoration, scene re- construction, and motion estimation. The classical problem in computer vision includes determining what the kind of object is in an image; this branch of computer vision is called object classification. In object classification, the system is given an image as input. The system should identify the label of the object to which the object belongs to. However, it is well known that even the best object classification algorithms will produce poor results when given poor features to track. Here, in literature survey made, different feature extraction techniques, many clus- tering algorithms, classification techniques and different approaches/methodologies are studied. Focus is restricted to these methods: Bag-of-Words (BoW) model and Convo- lution Neural Network. Here, Experiments that are performed on BoW model are implemented using Mi- crosoft Visual Studio with OpenCV libraries. The BOW model extracts the SIFT and SURF features from all the training images. These features are clustered using the k- means to create the dictionary of visual words. Next, SVMs with linear and RBF kernel are used then for the classification purpose. Task is also addressed through a well known convolution neural network - ALEXNET. The thesis also proposes variants of ALEXNET. Results are compared with state-of-the-art and proves the effectiveness of the proposed models.