Variability Distortion Research Articles

H.264 Coarse Grain Scalable (CGS) and Medium Grain Scalable (MGS) Encoded Video: A Trace Based Traffic and Quality Evaluation

The scalable video coding (SVC) extension of the H.264/AVC video coding standard provides two mechanisms, namely coarse grain scalability (CGS) and medium grain scalability (MGS), for quality scalable video encoding, which varies the fidelity (signal-to-noise ratio) of the encoded video stream. As H.264/AVC and its SVC extension are expected to become widely adopted for the network transport of video, it is important to thoroughly study their network traffic characteristics, including the bit rate variability. In this paper, we report on a large-scale study of the rate-distortion (RD) and rate variability-distortion (VD) characteristics of CGS and MGS. We found that CGS achieves low bit rate overheads in the 10–30% range compared to H.264 SVC single-layer encodings only for encodings with a total of up to three quality levels; more quality levels result in substantially higher overheads. The traffic variabilities of CGS are generally lower than for single-layer streams. We found that in the low to mid range of the MGS quality scalability, MGS can achieve the same or even slightly higher RD efficiency than corresponding single-layer encoding; toward the upper end of the MGS quality scalability range the RD efficiency drops off significantly. MGS layer extraction following the hierarchical B frame structure gives nearly as high RD performance as RD-optimized extraction. In the range of high RD efficiency, MGS streams have significantly higher traffic variabilities than single-layer streams at the frame time scale. At the group-of-pictures (GoP) time scale, MGS has similar or lower levels of traffic variability compared to single-layer streams. Generally, MGS layer extraction over the time horizon of individual GoPs gives significantly lower traffic variability than extraction over the time horizon of the full video sequence.

Corrections to Video Texture and Motion Based Modeling of Rate Variability-Distortion (VD) Curves

In the above-named article, due to a production oversight, mistakes were introduced prior to final printing. MPEG-4 appeared as -4 throughout, and should have appeared as MPEG-4. In [1, p. 645, Table VI], the caption should have read: COMPARISON OF Jmax ESTIMATES BASED ON PIECEWISE APPROXIMATION (P) WITH 2, 3, AND 4 SAMPLES AND COMBINED MODEL (RRMSE (%) VALUES) FOR PERMITTED LOSS PROBABILITIES OF e = 10-7 AND 10-3. In [1, sec. V, p. 645], the last sentence of the first paragraph should have read: From these considerations, we can already formulate two high-level recommendations for the choice of q1 and q2: (i) there should be a reasonable interval size Δq = q1-q2, and (ii) at least one q should be chosen in the high compression ratio region where the hump is typically situated. In [1, p. 647], -4 should have read as MPEG-4 in references [7], [14], and [15]. In reference [17], -encoded should have read MPEG-encoded. In reference [26], -2 should have read MPEG-2. In reference [28], all-IP? was omitted. In reference [32], -7 should have read MPEG-7. The corrected references [7], [14], [15], [17], [26], [28], and [32] are provided.

Corrections to "Video texture and motion based modeling of rate variability-distortion (VD) curves"

In the above-named article, due to a production oversight, mistakes were introduced prior to final printing. "MPEG-4" appeared as "-4" throughout, and should have appeared as "MPEG-4". In [1, p. 645, Table VI], the caption should have read: COMPARISON OF J <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">max</sub> ESTIMATES BASED ON PIECEWISE APPROXIMATION (P) WITH 2, 3, AND 4 SAMPLES AND COMBINED MODEL (RRMSE (%) VALUES) FOR PERMITTED LOSS PROBABILITIES OF ε = 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-7</sup> AND 10 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-3</sup> . In [1, sec. V, p. 645], the last sentence of the first paragraph should have read: "From these considerations, we can already formulate two high-level recommendations for the choice of q1 and q2: (i) there should be a reasonable interval size Δq = q1-q2, and (ii) at least one q should be chosen in the high compression ratio region where the "hump" is typically situated." In [1, p. 647], "-4" should have read as "MPEG-4" in references [7], [14], and [15]. In reference [17], "-encoded" should have read "MPEG-encoded." In reference [26], "-2" should have read "MPEG-2." In reference [28], "all-IP' was omitted. In reference [32], "-7" should have read "MPEG-7." The corrected references [7], [14], [15], [17], [26], [28], and [32] are provided.

Video Texture and Motion Based Modeling of Rate Variability-Distortion (VD) Curves

We examine and model the bit rate variability-distortion (VD) curve of -4 part 2 variable bit rate (VBR) video encodings. The VD curve has important applications for evaluating the statistical multiplexing of streaming video. We show that the concave VD curve shape at high compression ratios, or equivalently large quantization scales, is influenced by both the texture and the motion information. Based on this insight, we first develop a general VD curve model by analytically expressing the VD curve in terms of elementary statistics (mean, variance, covariance) of the numbers of motion and texture coding bits. In a second step we develop and validate linear and quadratic models for the elementary texture and motion bit statistics, whereby the model parameters are obtained from only two sample encodings. The texture and motion bit models are then employed in our general VD curve model. This work extends our previous work on piecewise models of the VD curve. The texture and motion based VD model obtained from two sample encodings has comparable accuracy to a piecewise model based on three sample encodings. In addition, the texture and motion based VD model provides fundamental insights into how texture and motion affect traffic variability.

The Rate Variability-Distortion (VD) Curve of Encoded Video and Its Impact on Statistical Multiplexing

Encoded video is expected to contribute a significant portion of the load on future communication systems and networks, which often employ statistical multiplexing. In such systems, the number of video streams that can be supported depends both on the mean bit rate as well as bit rate variability of the video streams. At the same time, the utility (revenue) earned from video streaming depends both on the number of supported video streams as well as their quality level. In this paper we examine the interplay between video quality, traffic variability, and utility for open-loop encoded video. We introduce the rate variability-distortion (VD) curve which relates the bit rate variability to the quality level of an encoded video. We find that the VD curve generally exhibits a characteristic hump behavior of first increasing, peaking, and subsequently decreasing variability for increasing quality. We examine the impact of video content characteristics, encoding parameters, and traffic smoothing on the VD behavior. We describe a methodology for assessing (i) the set of the video streams that can be supported with a statistical quality of service requirement, and (ii) the utility earned from video streaming over a link. This methodology is based on the rate-distortion and rate variability-distortion characteristics of the videos. We find that the statistical multiplexing gain and the utility as a function of the video quality level typically exhibit a hump similar to the VD curve.