Research Commons
      • Browse 
        • Communities & Collections
        • Titles
        • Authors
        • By Issue Date
        • Subjects
        • Types
        • Series
      • Help 
        • About
        • Collection Policy
        • OA Mandate Guidelines
        • Guidelines FAQ
        • Contact Us
      • My Account 
        • Sign In
        • Register
      View Item 
      •   Research Commons
      • University of Waikato Theses
      • Higher Degree Theses
      • View Item
      •   Research Commons
      • University of Waikato Theses
      • Higher Degree Theses
      • View Item
      JavaScript is disabled for your browser. Some features of this site may not work without it.

      Improving Bags-of-Words model for object categorization

      Zhang, Edmond Yiwen
      Thumbnail
      Files
      thesis.pdf
      12.42Mb
      Citation
      Export citation
      Zhang, E. Y. (2013). Improving Bags-of-Words model for object categorization (Thesis, Doctor of Philosophy (PhD)). University of Waikato, Hamilton, New Zealand. Retrieved from https://hdl.handle.net/10289/7487
      Permanent Research Commons link: https://hdl.handle.net/10289/7487
      Abstract
      In the past decade, Bags-of-Words (BOW) models have become popular for the task of object recognition, owing to their good performance and simplicity. Some of the most effective recent methods for computer-based object recognition work by detecting and extracting local image features, before quantizing them according to a codebook rule such as k-means clustering, and classifying these with conventional classifiers such as Support Vector Machines and Naive Bayes.

      In this thesis, a Spatial Object Recognition Framework is presented that consists of the four main contributions of the research.

      The first contribution, frequent keypoint pattern discovery, works by combining pairs and triplets of frequent keypoints in order to discover intermediate representations for object classes. Based on the same frequent keypoints principle, algorithms for locating the region-of-interest in training images is then discussed.

      Extensions to the successful Spatial Pyramid Matching scheme, in order to better capture spatial relationships, are then proposed. The pairs frequency histogram and shapes frequency histogram work by capturing more redefined spatial information between local image features.

      Finally, alternative techniques to Spatial Pyramid Matching for capturing spatial information are presented. The proposed techniques, variations of binned log-polar histograms, divides the image into grids of different scale and different orientation. Thus captures the distribution of image features both in distance and orientation explicitly.

      Evaluations on the framework are focused on several recent and popular datasets, including image retrieval, object recognition, and object categorization. Overall, while the effectiveness of the framework is limited in some of the datasets, the proposed contributions are nevertheless powerful improvements of the BOW model.
      Date
      2013
      Type
      Thesis
      Degree Name
      Doctor of Philosophy (PhD)
      Supervisors
      Mayo, Michael
      Publisher
      University of Waikato
      Rights
      All items in Research Commons are provided for private study and research purposes and are protected by copyright with all rights reserved unless otherwise indicated.
      Collections
      • Higher Degree Theses [1714]
      Show full item record  

      Usage

      Downloads, last 12 months
      33
       
       

      Usage Statistics

      For this itemFor all of Research Commons

      The University of Waikato - Te Whare Wānanga o WaikatoFeedback and RequestsCopyright and Legal Statement