The present invention provides a deep learning method-based block segmentation coding complexity optimization method and device, the method comprising: in an HEVC, checking a frame coding mode currently used in the HEVC; selecting a CU segmentation prediction model corresponding to the frame coding mode according to the frame coding mode; the CU segmentation prediction model being a pre-established and trained model; predicting a CU segmentation result in the HEVC according to the selected CU segmentation prediction model, and segmenting the entire CTU according to the predicted CU segmentation result. In a particular application, if the frame coding mode is an intra-frame mode, then the CU segmentation prediction model is an ETH-CNN which can be terminated in advance; and if the frame coding mode is an inter-frame mode, then the CU segmentation prediction model is an ETH-LSTM which can be terminated in advance and the ETH-CNN. Said method significantly shortens the time required for determining the CU segmentation during coding in the premise of guaranteeing the CU segmentation prediction precision, effectively reducing HEVC coding complexity.
展开▼