Comparisons and Improvements of R-D Performance of Error Resilient Techniques for Robust Image/Video Transmission over Error-prone Networks

• Main Authors: Kuo-liang Hsu, 許國良
• Other Authors: Day-fann Shen
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/86225295129605915059

碩士 === 國立雲林科技大學 === 電機工程系碩士班 === 94 === The representation of image and video signal requires large amount of data, which must be compressed for efficient transmission over the network. However, the compressed image/video data is very vulnerable to network errors namely noises and packet loss. Thus, error resilient technique is essential for image/video communication over error prone network. Many source coding based error-resilient techniques have been proposed, however, studies in performance comparison among different categories of techniques are not seen so far. In this thesis, we are interested in comparing the Rate-Distortion performance between two different error-resilient technique categories namely data hiding and Multiple Description Coding (MDC); In addition a simple and novel source coding based error-resilient technique called Redundant Sub-sampled I frame (RSI) is also proposed for comparison in R-D performance. Before the comparison, we firstly derive a reversible data embedding technique called 2C1B for transformed coefficients based on the difference expansion; we then propose two novel difference expansion based reversible data hiding techniques namely 3C2B and 1C1B with data hiding capacity of 133% and 200% respectively with respect to the 2C1B method. These three reversible data hiding techniques are then compared with ODD-EVEN irreversible data hiding technique for JPEG image transmission. The experimental results show that the 1C1B reversible data embedding technique outperforms the other three in R-D performance in error condition. We improve the R-D performance further by modifying the JPEG Huffman Table reflecting the changes due to 1C1B embedding. Secondly, we proposed the WS/DDC (Weighted Sub-sampling/Double Description Coding, MDC category) and RSI (Redundant Sub=sampled I frame, a simple and novel error=resilient technique) to be compared with the 1C1B, the best in the data embedding category. Experiments show that RSI performs the best in R-D sense for packet loss rate 10% and below; the WS/DDS performs the best for packet loss rate 10% and above. In all cases, the 1C1B data embedding performs the worst. Finally, we are interested in comparing the R-D performance between FEC channel coding and the source coding based error-resilient RSI as well as DDC. Experimental results show that for noises in the forms of random or burst bit errors in physical layer, the FEC channel coding adopted in IEEE802.16a outperforms significantly the source coding based DDS and RSI. The source coding based DDS or RSI are mainly for packet losses due to network congestions.