Summary: | 碩士 === 國立臺灣大學 === 資訊工程學研究所 === 99 === Multimedia is the computer information that combines several communication media such as audio, video, text, etc. With the growing up of Internet, people are used to acquire information in electronic formats instead of traditional forms. To increase the capacity for current communication channel, data compression is a necessary tool.
Compression is a method to reduce the file size of source data. The source data could be in various kinds of formats. Since we are living in a world full of multimedia information now, the most popular forms of multimedia message are images or videos. In the past few years, many image (video) compression standards had been developed to reduce the size of source data. Hence we could carry more information on a transmission channel with fixed bandwidth.
One of the most important techniques in image (video) compression is variable length coding. Compared with fix length code, the variable length coding scheme requires smaller number of bits to represent a given message.
In this thesis, we survey the role of variable length code in modern codec. Then in the first part, we propose a new data structure which facilitates to decode multiple symbols in a lookup step. The experimental result shows it could speed up the decoding process efficiently.
In the second part, we study the relationship between zigzag scan and variable length coding. Since the result of zigzag scan decides the lowest entropy of the run-length pairs so as to the efficiency of the variable length coding hereafter, we could further improve the compression ratio by choosing customized zigzag patterns for specific images (videos). As the result, we develop a new zigzag scanning scheme for this purpose. Experimental results show we could reduce at most 1% with only little computational cost.
|