Adaptive Packet Significance Classification for Multiple H.264 Video Streams over Networks

Abstract

With technology advances in multimedia compression and Internet development, video streaming applications are full of potentials. By utilizing the current video coding techniques such as H.264, the significances of video packets in different video sequences may differ significantly. Equal error protection to all video packets regardless the significance may seriously degrade the quality of video sequences with high motion, complex texture, or intra-coded.

Therefore, this thesis proposes a packet significance classification mechanism for multiple H.264 video streams over networks. This framework consists of two main parts. First, we propose an adaptive packet classification mechanism in the application layer. Based on the packet location in video frames, error propagation effect, and video sequence complexity, this mechanism can adaptively estimate the significance of each video packet. Second, we propose an adaptive QoS mapping mechanism based on video classes for multiple video users over networks. In contrast to traditional QoS mapping mechanisms that do not consider the differences of video properties among users, this study takes multiple video users into consideration. Different video sequences belonging to various users are assigned different QoS mapping criteria to the limited service levels in the network layer.

Comparing the overall video transmission efficiency with traditional methods in simulations, the received video quality belonging to the users that send complex video sequences can be improved up to 6.67 dB. Although the received quality of video with low motions is slightly degraded, the overall received quality is still acceptable.