The HEVC (High Efficiency Video Coding) codec has passed its headline target of 50% reduction in bitrate for a given video quality compared with its MPEG 4/AVC/H.264 predecessor in only the first version. This has emerged from the MPEG standards group, which has just completed the most extensive testing so far of HEVC Version 1 against a wide range of video material encoded at varying qualities and resolutions following the standard’s finalization last year.
Halving the bit rate was also the target for the preceding H.264, but in that case it took over 10 years of optimization to pass the mark. Introducing a new encoding standard is an expensive exercise requiring substantial algorithmic and coding development as well as replacement of infrastructure across the video lifecycle and so far has come in approximately 10 year cycles. Within those cycles the existing encoding scheme is progressively optimized to yield improvements in compression ratio, quality and performance, until a point is reached when further investment yields such diminished returns as to be counterproductive. New requirements for picture quality emerge that cannot be sustained with the current system, while at the same time computational advances cease to be fully exploitable.
In the case of HEVC there was a need to cater properly for the capabilities of emerging screen sizes and of course the Ultra HD format, which meant supporting 10 bit color and higher frame rates. On the processing front it meant being able to exploit the parallel processing capabilities of generally available multicore chips and so be better placed to run in software rather than dedicated ASICs, as well as enabling efficient decoding on hand-held devices without consuming too much power.
Its power efficiency has yet to be fully proven, but the recent MPEG tests have put the first version through its paces in many other respects. They have gone some way to prove our own early prediction at Faultline that HEVC would end up surpassing the 50% benchmark for compression performance by a considerable margin in the case of UHD content at least. At its 108th meeting in April in Valencia, Spain, MPEG concluded that ‘clear evidence was found that HEVC is able to achieve 50% bitrate savings and more, compared to the AVC High Profile’. MPEG went on to say that the precise results would be made available publicly in the report N14420, to be published on the MPEG website.
MPEG also pointed out that HEVC was still in a state of development with further improvements to come in the second version scheduled for completion by the end of 2015, beyond the optimizations that have just been achieved for the recent testing. These extensions include Screen Content Coding, specifying compression of video containing rendered graphics and animation. It also includes the ‘Range Extensions amendment’, introducing technology for more efficient compression of video content for color sampling formats beyond 4:2:0 and up to 16 bits of processing precision.
Fundamentally HEVC has been designed from the ground up to handle high resolutions more efficiently and also with better performance through exploitation of parallelization. Firstly the fundamental unit of compression, the macroblock, was replaced by larger and more flexible Coding Tree Units (CTUs), which can now comprise up to 64×64 pixels, or 4096 altogether, compared with 16×16 or 256 altogether before.
This allows greater scope for breaking the picture into different sized elements to match variations across the image, so that some can be just 4×4, others 16×16, yet others 32×32 and so on. Then at a higher level of structure HEVC introduced tools for parallel processing of multiple CTUs, including tiles and wave fronts. Tiles are rectangular blocks of CTUs that can be encoded and decoded independently of each other with the main purpose being to allow frames to be resynchronized in the event of data loss, therefore contributing to resiliency.
Wavefronts on the other hand entail dividing a slice of video into successive rows of CTUs each of which is dependent on encoding decisions made in the preceding row. The effect is to allow encoding to proceed efficiently across a frame like an expanding wave front, but requiring help in the case of data loss.
There is of course far more to HEVC, but the basic idea is to allow for parallelism in encoding and decoding, along with greater flexibility in partitioning frames up for the processing to take account of the internal picture structure. This flexibility becomes increasingly advantageous at higher resolutions when there are more pixels per image and is why HEVC achieves greatest levels of compression efficiency and performance for UHD.