Nowadays, the content growth of HD and the HD broadcasting service are offering users to enjoy high quality and high resolution. In the near future, the content growth of UHD and the UHD broadcasting service also will be offered to users following the demands to needs for higher quality and higher resolution. For these high quality and high resolution video, the HEVC standard, the next generation of video coding standard, was developed recently on the JCTVC team. This HEVC was aimed to have two times higher coding efficiency than H.264/AVC, and now HEVC has about 35% data compression ratio when compared to the H.264/AVC using HEVC Test Model (HM) 9.0.
The coding structure of the HEVC is based on the block based hybrid video coding structure by using the motion estimation/motion compensation (ME/MC) and discrete cosine transform (DCT) but it is much more sophisticated than the previous video coding standards. The fixed-size ME for the macroblock (MB) is changed into the variable-size prediction unit (PU) ME for the variable-size coding unit (CU). The H.264/AVC first introduced the variable-size ME varying from 16×16 down to 4×4 within the 16×16 MB and then achieved high video coding gain. However, the 16×16 block size is too small when applied to high resolution video such as 4k×2k and 8k×4k. For these high quality and resolution video sequences, the HEVC adapted a recursive quad-tree structured CU, which significantly improved coding efficiency but leaded to dramatic increase of the computational complexity.
To reduce the computational complexity of the HEVC encoder, the fast CU size decision algorithms for the fast encoding of HEVC are proposed in this paper. First, we propose a fast CU decision algorithm based on initial CU estimation and early PU termination by using the CBF and RD cost. The HEVC encoder tests almost every CU for finding the best CU combination, which begins with LCU and ends with the smallest CU. This unconditional encoding start with LCU tends to waste the encoding time. Thus, we predict the initial CU depth effectively by using spatial and temporal characteristics of the video sequence before LCU encoding. Therefore, unnecessary large size CU encoding could be skipped and we can reduce the encoding time. In addition, we skipped the remaining PU process by early PU termination algorithm. We check and save the CBF and RD cost for each Inter PU mode. If these CBF and RD cost are satisfied with our condition, any remaining PU process are terminated within the current CU. In addition, encoding of sub-CUs of the CU is omitted. If early Inter PU termination is not occurred during the process of Inter PU modes, we apply Intra PU skip algorithm based on sub-CBF values obtained in Inter PU modes. This proposed algorithm achieves significant encoding time saving. In random access coding, The average encoding time savings are 37.97%, 49.91 when the weighted factor α value is 1.0 and 1.5 respectively. In low-delay P coding, The average encoding time savings are 33.61%, 43.52% depending on the weighted factor α value.
we also propose the fast CU size decision based on hierarchical layer in temporal prediction structure of HEVC. This algorithm use the frame’s visual quality change depending on the hierarchical layers in random access structure which is one of the temporal prediction structures in HEVC to reduce the computational complexity in CU decision procedure. The structure of random access has four hierarchical layers according to the frame’s QP. Thus, the frames in this structure have the hierarchical visual quality change depending on layer. The proposed algorithm uses this characteristic of the hierarchical layer to select the reference in the previously encoded frame for estimating the current CU size early. And this algorithm performs the CU encoding from 64×64 CU to the predicted CU size. And also, In order to increase the rate of CU size decision more, the method using the distortion of SKIP mode is grafted onto the proposed algorithm. The proposed algorithm significantly reduce computational complexity of the HEVC encoder by up to 37.39% with only a small PSNR loss and bit rate increment.