Designing Efficient Parallel for Some Digital Image Processing Problems

• Main Authors: Yu-Hua Lee, 李鈺華
• Other Authors: Shi-Jinn Horng
• Format: Others
• Language: en_US
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/79884354604782866127

博士 === 國立臺灣科技大學 === 電機工程系 === 91 === Image processing is the science to the analysis and manipulation of pictorial information. Generally speaking, the pictorial information means two-dimensional or three-dimensional visual image. Any operation that acts to improve, correct, analyze, or change an image in some way is called image processing. Digital image processing uses digital circuits, computer processors, and software to carry out an operation. Digital techniques provide the precise implementation of image processing function, along with a great flexibility and power for general-purpose image processing applications. Digital image processing has become a significant form of image processing of continuing improvement in sophisticated semiconductor technologies. Digital image processing algorithm research also continues as a fast pace, spurred on by increasing commercial demands. When coupled with an expanding application base, digital image processing techniques are becoming key tools in diverse new industries. In this thesis, we focus our attention on the digital image processing problems. In recently years, lots of image processing applications require real-time processing or large image data processing rate within a limited time. So, in this thesis, we focus on applying the massively parallel processing techniquess to solve the high computational complexity of digital image processing problems, most of these problems can be solved efficiently on the massively parallel processing models. We focus our attention on sev-eral intermediate-level image processing problems including two-dimensional chessboard distance transform (2D-CDT) problem, two-dimensional medial axis transform(2D-MAT) problem, and three-dimensional Euclidean distance transform (3D-EDT) problem.