Super-Resolution with Sparse Representation by Massive Parallel Computing
| Autor: | Hao-RongDing, 丁浩容 |
|---|---|
| Rok vydání: | 2014 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 102 This thesis proposes a parallel computing design algorithm for image Super Resolution with sparse representation. To reduce the algorithm’s processing time, we use the Computing Unified Device Architecture based on the concept of parallel computing. With the consideration on hardware’s limitation, we modified the original algorithm’s computing steps to achieve the goal for decreasing the processing time. In parallel computing design algorithm, we consider the least absolute shrinkage and selection operator (LASSO) method to speed up the processing time in finding the sparse coefficient. The proposed method is easy for operating on the GPU hardware computing. Compare with the original method implemented on Matlab, the GPU based algorithm can achieve 7.3 times faster than original through the parallel computing design concept. We replace the computing for sparse coefficient through the method of modified LASSO and the processing time can get 13.9 faster than original’s execution time. For the image quality, we adapt some image quality assessment standard such as the Pick Signal to Noise Ratio (PSNR), the Feature Similarity (FSIM), and the Structure Similarity (SSIM). |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|