SMPTE 2014: HEVC Bitrate Savings Verified

Codec put to the test
Author:
Publish date:
Social count:
0

HOLLYWOOD—The theory of high-efficiency video coding is that it cuts the bitrate of H.264 by half. John Pallett of Telestream and his team went looking for what it could really do.

Pallett said he previously participated in a study done on x265 0.7—an open source version of HEVC—that indicated a 25 to 30 percent bitrate savings. The same experiments were done again with updated technology.

To determine an HEVC bit-rate recommendation, they set a quality target based on MPEG-2 and H.264 SD and HD. They identified HEVC equivalent bitrates to achieve that quality target, and used those quality targets to identify an ultra HD bitrate. The goal was to identify a baseline.

Structural similarity index, or SSIM, and peak signal-to-noise ratio were used as primary metrics, and did not take into account overall motion. SSIM does a “fairly good job of capturing quality differences between codecs,” Pallett said.

Both average and worst frame results were measured. High motion footage was used—“Tears of Steel,” and “Extreme Sports,” provided by Dolby and Brain Farm. Both were downscaled from ultra HD, to HD to SD. Both were 8-bit, 24fps, so there was no messing with interlace, he said.

Quality target was SSIM (avg) .977 for SD at 860 Kbps, and .975 for HD at 2.7 Mbps. For ultra HD, nearly 12 Mbps was necessary for the projected quality level.

“We came to a conclusion of a 49 percent bitrate savings for SD and HD,” Pallett said.

In a study earlier this year, Pallett said it was observed that compressed frame sizes appeared to be more variable with HEVC compared to H.264. Therefore, a larger buffer window is probably a good idea because there is less predictability of what’s going on. Pallett and Co. measured the effect of buffer size on video quality using the video buffering verifier model.

“There was a noticeable increase in quality at around the two-second mark for H.264,” he said.

Running several hundred encodes, Pallett said the average quality frames were about the same, but the worst frames improved at two seconds VBV. Buffer size also affects HEVC more than H.264, he said.

Pallett said he and his colleagues concluded that x265. 1.2 is getting close to the theoretical numbers. Bitrates savings of 40 to 45 percent compared to H.264 are achievable in real-world applications, he said.