March 11, 2016 -- Nowadays HEVC (H.265) decoder settled down in video codec industry but HEVC encoder is not yet.
Guarantee for overwhelming encoding efficiency comparison with H.264, HEVC encoding engine must cover many and various complex encoding tools.
Specially to support every transform size from 32×32 to 4×4, it is required huge computing resource.
But the improvement of micro-semiconductor fab process is slowing down and the rising of performance is really not easy.
Nonetheless, IT industry needs more computing power to process high resolution video for HDR, VR, AR, hologram and other high-end technic.
Also, IT industry wants low power solution for mobile, wearable and IoT application.
MnHEVC overcome the barrier of optimisation with deep learning acceleration and 8-bit anti-banding encoding technic.
MnHEVC is deep learned with many and various video estimation patterns and its depth classifier make to reduce encoding complexity dramatically.
MnHEVC-DeepL classify transform size (every size from 32×32 to 4×4) deep-learned from fully depth estimation encoder engine.
MnHEVC-DeepL is accelerated encoding performance as 22.7%, but use full encoding tools and keep same video quality as original MnHEVC encoder.
MnHEVC-DeepL encode up to 4K 30fps at 395Mhz or 4K 45fps (1080P 30fps 6-channel for VR encoding) at 600Mhz only with 6,000K gate count included memory.
The other things to reduce resource, MnHEVC support 8-bit pixel anti-banding encoding.
Generally, when encoding 4K image, 8-bit pixel encoding occur banding effect, because of disappearing low frequency gradient.
For reducing banding effect, it needs 10-bit or 12-bit pixel encoding but that system make rising SoC resource.
MnHEVC IP estimate image pattern and decide encoding mode for reducing banding effect.
4K 8-bit pixel raw image
4K 8-bit pixel normal encoding (10.428Mbps)
4K 8-bit anti-banding encoding (10.428Mbps)
MnHEVC 8-bit ant-banding encoding make to reduce SoC complexity, bus bandwidth, video IP core area and bit rate increasing on 10-bit based encoding.
MnHEVC deep learning acceleration will be available at 2016 April and 8-bit anti-banding encoding is stable now.
Mn_nH will provide you with high performance compact HEVC encoder IP and continue to optimise highly with plentiful inspiration.
