Loading...

Localized Multiple Kernel Learning for Image Classification

Zamani, Fatemeh | 2018

2267 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 51408 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Jamzad, Mansour
  7. Abstract:
  8. It is not possible to compute a linear classifier to classify real world images, which are the focus of this thesis. Therefore, the space of such images is considered as a complex. In such cases, kernel trick in which data samples are implicitly mapped to a higher dimension space, leads to a more accurate classifier in such spaces. In kernel learning methods, the best kernel is trained for the classification problem in hand. Multiple Kernel Learning is a framework which uses weighted sum of multiple kernels. This framework achieves good accuracy in image classification since it allows describing images via various features. In the image input space which is composed of different extracted feature vectors there are images from the same class which are dissimilar (intra class variance) and there are images from different classes which are similar (inter class relationship). In the Multiple Kernel Learning framework, fixed weights are computed for each kernel which is not appropriate to classify images because of the mentioned complexities. The goal of this thesis is to present a kernel learning based approach to classify images by computing localized weights for kernels. Due to the power of SVM in data classification, an algorithm is designed based on SVM in which variable weights for kernels are computed based on the gating functions. Moreover, because of the successfulness of Sparse Representation based Classifier (SRC), a new version of SRC is proposed based on multiple kernels. In the proposed algorithm, a multiple kernel dictionary is trained. To this end, kernel weights are embedded as new variables to the goal function of sparse representation optimization problem. The convexity of the sub-problem which is composed to compute kernel weights is proved. To evaluate the proposed algorithms, after extracting SPM histograms which describe images, some experiments are conducted on the two challenging image datasets, Caltech 101 and Flower 17. The results demonstrate high performance of the proposed algorithms comparing to the state of the art
  9. Keywords:
  10. Images Classification ; Dictionary Learning ; Multiple Kernel Learning ; Lecalized Kernel Weighing ; Sparse Representation Based Classifier ; Feature Fusion

 Digital Object List

 Bookmark

...see more