Loading...

Sparse Representation and Dictionary Learning based Methods for Skin lesion Segmentation and Classification

Moradi Davijani, Nooshin | 2021

182 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 54735 (02)
  4. University: Sharif University of Technology
  5. Department: Mathematical Sciences
  6. Advisor(s): Mahdavi Amiri, Nezamoddin
  7. Abstract:
  8. Skin cancer is a common type of cancer in the world. Melanoma is considered as the deadliest form of human skin cancer that causes approximately 75% of deaths associated with this cancer. However, melanoma is curable if diagnosed in an early stage. Due to high visual similarities and diverse characteristics of lesions, it is a challenging task to differentiate between different types of skin lesions. Therefore, it is worthwhile to develop a reliable automatic system increasing the accuracy and efficiency of pathologists. Here, we propose sparse representation and dictionary learning based methods for skin lesion segmentation and classification. First, we review the steps of a computer aided diagnosis (CAD) system and available models and algorithms for sparse representation and dictionary learning. Then, we propose a kernel sparse representation method for binary segmentation and classification of skin lesion images, and finally adapt the proposed framework for multi-class classification. The main idea of our framework is based on a kernel sparse representation, which produces discriminative sparse codes to represent features in a high-dimensional feature space. After that,we propose a method based on joint dictionary learning for multi-class segmentation of dermascopy images. The key idea of our framework is based on combining data from different feature spaces to jointly learn two dictionaries. Finally, a graph-cut method is used to make the final segmentation based on both the topological information of skin lesions and the learned dictionaries. We implement the proposed methods and evaluate them on different skin lesion datasets. The comparison results with other available methods show the effectiveness of the proposed method for skin lesion image analysis
  9. Keywords:
  10. Sparse Representation ; Dictionary Learning ; Segmentation ; Classification ; Kernel Methods ; Skin Lesion Image Analysis ; Skin Cancer

 Digital Object List

 Bookmark

No TOC