Fast partition algorithm in depth map intra coding unit based on multi-deep convolution neural network

The three-dimension high-efficiency video coding standard (3D-HEVC) finalized comes with a significant increase in complexity caused by the integration of depth map coding technology. This complexity is primarily triggered by the quad-tree partition of the Intra Coding Units (CU) in the depth map. A new technique utilizing deep learning is proposed, in this paper, to tackle the issue of excessive complexity aiming to predict efficiently the CU partition structure. The proposed method involves building a dataset of CU partition structure information for a depth map, creating a Multi-Deep Convolutional Neural Network (MD-CNN) model using this dataset, and then incorporating the model into the 3D-HEVC test platform. This approach reduces the 3D-HEVC video encoder complexity by 48.29% without affecting robustness, compression efficiency and video quality.