Loading...

Adaptive Transform using Lifting Scheme with Applications to Object Detection in Video Images

Amiri, Mahdi | 2010

723 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 40387 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Rabiee, Hamid Reza
  7. Abstract:
  8. In this thesis, a novel method for the design of adaptive wavelet transforms based on the lifting scheme structure is presented. We have compared the proposed adaptive wavelet design method with the exisiting algorithms by providing a brief survey of the recent adaptive lifting scheme techniques. In addition, object detection is selected as the target application and two novel template matching algorithms based on the proposed adaptive lifting scheme transform, called LAPT and RASIM, are presented. As the main building block of the proposed algorithms, given an object as a template, we first select a set of coefficients as object features in the wavelet transform domain and then build an adaptive transform based on the selected features. The goal of the new adaptive transform is to vanish the selected features in the transform domain. After applying both non-adaptive and adaptive transforms to a given test image, the corresponding transform domain coefficients are compared for detecting the object of interest. In the LAPT algorithm, we have combined the core detection algorithm with proper log-polar mapping model in the parametric template space to attain rotation/scale invariance property. Having the proper set of vertex templates in LAPT, as most of the time consuming tasks are transferred to the offline steps, efficient object detection in the online steps will be performed with very low computational complexity. The second proposed template matching algorithm is called RASIM. It is a novel algorithm for matching image interest points. Potential interest points are identified by searching for local peaks in Difference-of-Gaussian (DoG) images. We refine and assign rotation, scale and location for each keypoint like the SIFT algorithm. Then, a pseudo log-polar sampling grid is applied to properly scaled image patch around each keypoint, and a weighted adaptive lifting scheme transform is designed for each ring of the log-polar grid. Our experiments show that the accuracy of RASIM is more than SIFT, which is the most widely used interest point matching algorithm in the literature. RASIM is also more robust to image deformations while its computation time is comparable to the SIFT. We have verified the properties of the proposed algorithms with comprehensive experimental results. Finally, the future work to further improve the performance of our new algorithms and the concluding remarks are presented.
  9. Keywords:
  10. Template Matching ; Object Recognition ; Adaptive Transform ; Lifting Scheme

 Digital Object List

 Bookmark

No TOC