Compressed video quality enhancement algorithm based on 3D-CNNs

By exploring the current block-based lossy video coding process and compressed videos, this paper finds two unique characteristics namely quality fluctuation and pixel deficiency. And we use 3D convolutional neural network (3D-CNN) to make full use of the limited temporal and spatial information in compressed video and build compressed video quality enhancement network (CVQENet) to improve the compressed video quality. The experimental results show that compared with the videos encoded by High Efficiency Video Coding (HEVC/H.265), the mean value of the Peak Signal-to-Noise Ratio (PSNR) of enhanced videos has been improved by 0.4652 dB under Low Delay (LD) configuration with Quantization Parameter (QP) is set to 37.