No Reference Quality Assessment For Networked Video Based On Bitstream

Quality assessment of networked videos is the key to ensure the quality of servicefor networked video applications. In this dissertation, the quality assessment fornetworked video is explored, and several quality assessment methods using thebitstream for networked video are proposed, including a packet-layer assessment model,two bitstream-layer assessment models and a frame type detection method forpacket-layer assessment models.The major contributions of this dissertation are summarized as follows:1. To improve the performance of packet-layer quality assessment of networkedvideos，a frame type detection method is proposed. Considering the characteristics ofthe compressed data of each type frame, dynamic thresholds are employed to roughlyestimate the type of each frame. Then, periodicity of the group of pictures (GOP) isused to polish the preliminary estimation. Finally, prediction structure of B-frames isdeterm ned us ng Spearman’s Rank orrelat on oeff c ent.2. To realize real-time and non-intrusive quality monitoring for networked videos,a packet-layer model for coding distortion assessment is proposed by considering themotion characteristic of video content. The frame type of each video frame isdetermined using the proposed frame type detection method. Then combining thecharacteristics of the bit-rate for coding I frames and P frames, an estimation of thetemporal complexity is proposed which reflects the motion characteristic of the videocontent. The proposed temporal complexity is incorporated in the bit-rate model,making it adaptive to different video content.3. To obtain more accurate quality of networked videos, a content-adaptivebitstream-layer (CABL) model is proposed for coding distortion assessment ofnetworked videos by analyzing the information extracted from packet headers andpayload of bitstreams. Firstly, the fundamental relationship between perceived codingdistortion and the quantization parameter (QP) is established. Then, considering the factthat the perceived coding distortion of a networked video significantly relies on both thespatial and temporal characteristics of video content, the spatial complexity is evaluatedusing the QP and the scale parameters of residual pixel distribution. Meanwhile, thetemporal complexity is obtained using the weighted motion vectors (MV). Finally, theproposed content-adaptive bitstream-layer model for coding distortion assessment isestablished by integrating these two content related factors into the fundamental relationship.4. To obtain more accurate video frame quality and provide the benchmarkedquality of video frame for quality assessment under packet loss, a coding distortionassessment based on frame quality is proposed for H.264/AVC networked video. Firstly,the relationship of the coding distortion of the video frame and quantization parameteris modeled. Then, the quantization parameters and bit-rates of I frames are employed topredict the spatial complexity of an I frame, and the motion vectors are employed topredict the temporal complexity of P frames. And then, the perceptual coding distortionof each frame is calculated taking account of the spatial and temporal masking effect ofthe human visual system，Finally, the video quality is obtained by pooling quality ofeach frame.