Manshor, Noridayu
(2013)
Fusion Of Global Shape And Local Features Using Multi Classifier Framework For Object Class Recognition.
PhD thesis, Universiti Sains Malaysia.
Abstract
Object class recognition deals with the classification of individual objects to a certain class. In images of natural scenes, objects appear in a variety of poses and scales, with or without occlusion. Object class recognition typically involves the extraction, processing and analysis of visual features such as color, shape, or texture from an object, and then associating a class label to it. In this thesis, global shape and local features are considered as discriminative features for object class recognition. For local features, misclassification problems occur if the object is too small and possess weak local features. Besides that, local features do not give implicit importance to the shape of the object, which is one of important features to human vision. Detecting objects is difficult if the pose changes. Consequently, pose changes will result in changes in shape features for an object in the same class. Hence, both local and shape features are combined in order to obtain better classification performance for each object class. Ultimately, a meta-classifier framework is proposed as a model for object class recognition. Meta-classifier is used to learn a meta-classifier that optimally predicts the correctness of classification of base classifier for each object. In this framework, individual classifiers are trained using the local and global shape features, respectively. Then, these classifiers results are combined as input to the meta-classifier. Experimental results have shown to be comparable, or superior to existing state-of- the-art works for object class recognition.
Actions (login required)
|
View Item |