Adaptive Packetization Of Video Stream Over Wireless Network

With the rapid development of the technologies of video coding and wirelesscommunication, multimedia communication over wireless networks achieves muchattention and has gained larger share of communication services in total. However, inreal video transmission applications, the challenge from two aspects should be welldealt with. On one hand, the original video data should be compressed efficiently tomeet the constraints on the limited-bandwidth of wireless channels. On the other hand,the use of variable length coding and prediction coding methods enhances thecorrelation of coded video data, hence the bitstream is highly sensitive to channel errors,i.e., bit errors or packet losses. Moreover, since wireless channel is time-varying andsubjected to interference, the wireless channel has high error bit rate and can notprovide the quality of service. Therefore, error control technologies to improve thequality of decoded video while efficiently exploiting channel bandwidth is an importantissue. In this thesis, the issues of error resilient video coding methods and unequalerror/loss protection schemes are studied. The major contributions of this thesis aresummarized as follows:1. The video frames with different types in a group of pictures (GOP) are unequalin their importance to the entire reconstructed video quality of the GOP. On the otherhand, different data partitions belonging to one slice have different contribution to thereconstructed video quality of the current slice. Jointly considering these two aspects, ametric to quantify the unequal importance of different data partitions in a GOP to thedecoded video quality is defined. Based on this metric, a priority sorting method forcoded video data coming from a GOP is proposed. Furthermore, according to thecharacteristics of wireless channels using orthogonal frequency division multiplexing(OFDM) technology, the video data marked as high priority are assigned to high qualitychannel to transmit; similarly, the video data with low priority are allocated to lowquality channel to deliver. Compared to the priority sorting method and transmissionscheme at frame level, the proposed GOP-level priority sorting method andtransmission scheme has better error resilience, thus can improve the quality of thedecoded video.2. Jointly considering the positions of different packets in a GOP and the contentcontained in different packets, a method to quantify the importance of packets ispresented. Furthermore, considering the syntax dependency among different datapartitions and the bursty packet loss property of wireless networks, an efficient unequal loss protection scheme exploiting channel coding is developed. In the proposed method,all the video packets with partition A and partial video packets with partitions B and Cin one GOP are provided more protection such that these important packets are arrivedwith low residual packet loss rate, thus improving the reconstructed video quality.3. Considering the fact that the macroblocks with different content characteristicshave different impact on the reconstructed video quality, a video coding method, whichtakes into account of content characteristics of macroblocks, is proposed for unequalloss protection. Jointly taking into consideration of spatial and temporal contentcharacteristics of each macroblock belonging to one video frame, a contentcharacteristics function is defined. In the proposed content characteristics function, twoadaptive weighting factors are employed to embody the spatial correlation between thecurrent macroblock and the adjacent macroblocks, and the temporal correlation betweenthe current macroblock and the co-located macroblock in the previous frame,respectively. Moreover, arrcoding to the coded type of the video frame to which themacroblock under consideration belongs, and the value of content characteristicfunction, the macroblocks in one video frame are arranged into the important slicegroup and the unimportant slice group to encode. Therefore, the slice group containingmacroblocks with complex texture or/and fast movement can be provided moreprotection using various unequal loss protection schemes to improve the reconstructedvideo quality.4. To overcome the disadvantages of high computational complexity and lowaccuracy of the existing methods for quantifying the relative importance of macrobloksin one video frame, the metric to measure the importance of macroblocks withlow-complexity and relative high accuracy are defined. In the proposed method, thecoded type of video frame and the content characteristics of macroblocks are jointlytaken into account. Furthermore, the error packet rate is related to the bit error rate andpacket length in wireless networks. Based on this fact, a flexible slice grouping methodto determine the content and length of each slice group adopting minimized channeldistortion metric is proposed.