Performance Evaluation of Image Engine on Multi-core Embedded Systems

• Main Authors: Shun-Mao Liao, 廖舜茂
• Other Authors: Cheng-Min Lin
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/07441739090845019086

碩士 === 南開科技大學 === 電子工程研究所 === 102 === MATLAB is the commercial mathematical software developed by MathWorks. MATLAB has a complete and powerful library on matrix operations, so it is widely used in the imaging algorithms, communication studies or numerical analysis. However, when MATLAB code is used to implement the research of the algorithm, most of the MATLAB codes still need to be translated into Java, C, or machine languages, especially in the codes need to implement on a hardware device. Hence, the FPGA or C language is considered. The algorithm has better performance without distorting the program semantics and excellent program structure after translation process. In this thesis, the cooperation of Chunghwa Picture Tubes (CPT), LTD. is happily obtained. The CPT provides four algorithms of image process including Super Clear, Wide Dynamic Range, White Balance, Multi-axis Color Adjust. These CPT algorithms are successfully translated into native codes running on Android embedded platform in our works. The program structure is adjusted and multiple threads are implemented to increase the performance of the algorithm. Finally, the experimental results show that when the number of threads equals to the number of CPU cores that obtain better execution performance. Using floating-point arithmetic algorithms have better performance if the central processor supports a floating-point arithmetic unit. In addition, the thesis proposed the automatic color measurement framework using in the panel production testing in the panel display validation and proofreading. The novel architecture is proposed to reduce the color style device for reducing the cost. The complexity of the operation is reduced when our proposed architecture.