bookmark: H265 vs H264 codec

A Comparison of H.264 and H.265

Function	H.264	H.265
Coding unit	16 × 16 macroblock	64 × 64, 32 × 32, 16 × 16 coding tree unit 64 × 64, 32 × 32, 16 × 16, 8 × 8 coding unit
Prediction	16 × 16, 16 × 8, 8 × 16, 8 × 8, 8 × 4, 4 × 8, 4 × 4	64 × 64 to 4 × 4, symmetric/asymmetric
Transform size	8 × 8, 4 × 4	32 × 32, 16 × 16, 8 × 8, 4 × 4
Transform	DCT	DCT, optional DST for 4 × 4
Intraprediction	9 modes	35 modes
Luma interpolation	6-tap filter for 1/2 sample followed by bilinear interpolation for 1/4 sample	8-tap filter for 1/2 sample, 7-tap filter for 1/4 sample
Chroma interpolation	Bilinear interpolation	4-tap filter for 1/8 sample
Interprediction	Motion vector	Advanced motion vector prediction (spatial and temporal)
Entropy coding	CABAC, CAVLC	CABAC
In-loop filtering	Deblocking	Deblocking followed by sample-adaptive offset
Parallel processing	Slices, slice groups	Slices, tiles, wavefronts

H265 GOP

Different from mpeg4, the B frame can used as referenced or unreferenced.

Unlike H264, H265 GOP could start without IDR frame.An IDR frame is independently coded and frames that follow it in the bitstream will not reference frames prior to it. To implement it, H265 define complex frame: 

• CRA: clean random access frame: new, independently coded frame that starts at an RAP
• BLA: broken-link access  frame
• RASL: random access skipped leading frame
• RADL:random access decodable leading frame

RASL frames and RADL frames are leading frames because their display order (i.e., encoder output order) precedes the RAP frame even though they appear after the RAP frame in the decoding order.

RDO (High Complexity) When Compared to No RDO

Mode	Encoding Time	Compression Efficiency	Video Quality
VBR	Longer (especially for low QP)	Lower	Better for every frame (especially for low QP)
CBR	Longer	NA	Better (especially for low QP)