Year: 2015
Type: dissertation
Abstract: Categorization is a fundamental problem of many computer vision applications, e.g., image \nclassification, pedestrian detection and face recognition. The robustness of a categorization \nsystem heavily relies on the quality of features, by which data are represented. The prior \narts of feature extraction can be concluded in different levels, which, in a bottom up order, \nare low level features (e.g., pixels and gradients) and middle/high-level features (e.g., the \nBoW model and sparse coding). Low level features can be directly extracted from images \nor videos, while middle/high-level features are constructed upon low-level features, and are \ndesigned to enhance the capability of categorization systems based on different considerations \n(e.g., guaranteeing the domain-invariance and improving the discriminative power). \nThis thesis focuses on the study of visual feature learning. Challenges that remain in designing \nvisual features lie in intra-class variation, occlusions, illumination and view-point \nchanges and insufficient prior knowledge. To address these challenges, I present several \nvisual feature learning methods, where these methods cover the following sub-topics: (i) \nI start by introducing a segmentation-based object recognition system. (ii) When training \ndata are insufficient, I seek data from other resources, which include images or videos in a \ndifferent domain, actions captured from a different viewpoint and information in a different \nmedia form. In order to appropriately transfer such resources into the target categorization \nsystem, four transfer learning-based feature learning methods are presented in this section, \nwhere both cross-view, cross-domain and cross-modality scenarios are addressed accordingly. \n(iii) Finally, I present a random-forest based feature fusion method for multi-view \naction recognition.
Author: Fan Zhu
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W1558113362