HEVC is the latest codec for video applications and is the successor to H.264 and MPEG-2 in the broadcast environment – especially for distribution to the home via a variety of methods. At the SMPTE Fall Conference, we listened to a paper from John Pallett at Telestream, which sought to compare the theoretical benefits of HEVC with real world encoding tasks against the results for H.264 and MPEG-2.
He did hundreds of encodes to see the difference that various parameters had on the results – all in an attempt to find the causes of artifacts or performance issues. He likened this hunt to the detective game of Clue (Cluedo for European readers) – was it Mr. Green in the parlor with the long GOP?
A similar previous study had used the 0.7 version of HEVC, and found a 25-30% improvement over H.264. The new study used the 1.2 version of HEVC (H.265) and was designed to set a quality target for the encoded video, determining what bit rates would be needed to achieve this. SD and HD content at 24 fps, 4:2:0, 8-bit was encoded using MPEG-2 and H.264 to establish the bit rates needed to achieve the quality targets. The content was then encoded using HEVC to determine comparable bit rates. These results were used to anticipate the results for UHD content encoding as well.
Evaluation of the quality was determined using SSIM (structural similarity index method) and PSNR (peak signal to noise ratio) – two common evaluation techniques of video signals. Pallett said he relied more on the SSIM results, however. Challenging high motion content was used for the encodes (Tears of Steel and Extreme Sports). The tables below show the data rates for the reference SD and HD content using MPEG-2 and H.264, plus the saving achieved using HEVC.
Pallett then determined what he thought the bit rates should be for an equivalent performance if the content had been available in UHD resolution. He concluded 21.6 Mbps for H.264 and 12.0 Mbps for HEVC, which represents a 45% saving over H.264.
In further analysis of the results, he found that increasing the video buffer from 1 second to 2 seconds made a big difference in the results – especially for the worse frame cases. As a result, his recommendation for Video Buffering Verifier (VBV) is 23.9 Mb. He concluded by noting that HEVC is much better on the motion artifacts than H.264, but evaluations were not done in his testing.
It should also be noted that these rates apply to 8-bit, 4:2:0 UHD, 24 fps content – typical of what will be streaming over Netflix and other channels in the near term. Increasing the bit depth to 10 bits theoretically adds about 25% to the data rate but in practice will more likely be 15-20%. Moving from 24 to 60 fps theoretically adds a 250% increase in the data rates, but will also be less in practice.