No Reference Quality Assessment Of Videos Affected By Packet Losses

With the rapid development of IP networks and mobile communication, videos inIP networks and mobile communication network has attracted more and more attention.There is an increasing demand for accurate and effective quality assessment ofunderlying videos. Accurate video quality assessment is crucial to video codecdevelopment, network protocol planning, in-network quality monitoring, etc. This thesisdevelops several no reference objective quality metrics for videos impaired by packetlosses. Because transmission error is one of the main causes of quality degradations ofnetworked video, a deep investigation on impacts of different attributes of packet losseson perceptual video quality is performed. The thesis is mainly concerned with thedesign and implementation of video quality to improve the timing and effect. Based onthe observed relationships between perceptual video quality and various attributes ofpacket losses, e.g. the new features of H.264/AVC, the type frame of loss, concentrativedegree of packet loss ratio and the tolerance characteristics of HVS, we propose severalnovel video quality metrics. The major contributions are outlined as follows:1. The new features of H.264/AVC, e.g. multi-mode intra coding technology,multiple reference frames inter prediction technology and advanced loop filteringtechniques etc., aggravated the error propagation’s effect of video quality. A noreference video quality assessment method is proposed taking error concealmenttechniques into account. By analyzing the affection of packet loss on video qualities, theproposed method can assess video qualities affected by packet loss via considering theaffections of spatial-temporal error concealments, motion compensation, intra prediction,and in-loop deblocking filter. Experimental results demonstrate that the consistency ofthe assessed video qualities is consistent with that assessed by the mean squared errorwhich should be calculated by comparing the reconstructed videos with the originalvideos. To overcome the shortcoming of MSE, a new reduced reference video quality ofstructural similarity metric is proposed by making use of the proposed method.Experimental results demonstrate that the consistency of subjective video quality withthe acquired by the proposed methods.2. In order to meet the real-time and low cost requirements of video qualityassessment over IP networks, a novel no-reference video quality assessment model isproposed. The temporal complexity is calculated without decoding, and the effect ofdifferent type of frame loss on video quality is considered. In the proposed model, bothspatial and temporal error concealment artifacts of HVS, forgiveness effect of singleloss, frame types are taken into account. Video qualities could be assessed without delay by utilizing the proposed model. Experiment results show that the scores obtained bythe proposed model highly agree with the subjective assessment scores. Compared withthe video quality evaluation model in G.1070, an average performance gain of6.96%could be achieved by the proposed model in terms of relativity to subjective results.3. Based on the effect of concentrative degree of packet loss ratio and features ofpacket ratio on video quality, a novel no-reference video quality assessment model isproposed. Without decoding, the temporal complexity is calculated in terms of the effectof packet loss ratio on video quality. The concentrative degree of packet loss ratio iscalculated in compression domain to determine the effect of concentrative degree ofpacket loss ratio on video quality. Then, the proposed video quality assessment model ispresented by taken spatial and temporal error concealment artifacts of HVS, packet lossratio, and the degree of concentration packet loss ratio into account. The proposedmodel can assess video qualities without fully decoding networked video, thus reducesthe cost and time delay. Experimental results show that the scores obtained by theproposed model are highly consistent with the subjective assessment scores. Comparedwith the video quality evaluation model in G.1070, a gain of6.18%in the averageperformance is achieved by the proposed model in terms of relativity to subjectiveresults.4. When networked video is transmitted over error prone network, thereconstructed videos at the receiver side might suffer from impairments introduced bypacket losses, only the compressed video bit-stream is available for quality assessmentand complete or partial decoding is usually not desirable at the network nodes. Themethod without resorting to the original video or video decoding is suited for real-timenetworked video application, a novel no-reference video quality assessment model isproposed. By analyzing the affection of packet loss on video qualities, the proposedmethod can assess video qualities affected by packet loss via considering the errorpropagation, error position, forgiveness effect of human vision and temporal distortion.The proposed model can assess video qualities without fully decoding networked video,thus reduces the cost and time delay. Experimental results show that the scores obtainedby the proposed model are highly consistent with the subjective assessment scores.