结合卷积神经网络的HEVC帧内编码压缩改进算法

近年来，卷积网络深度学习已在图像处理、目标检测等领域取得巨大成功。受其启发，将卷积神经网络(CNN)应用于传统视频压缩标准已成为一个新的研究热点。本文提出一种集成卷积神经网络的高效视频编码(HEVC)压缩改进算法，将下采样过程、HEVC的编解码过程、上采样及质量增强过程集成为一体。为高效提取视频帧的结构特征，在所提压缩算法中集成了两个卷积神经网络。提出了一种下采CNN(DwSCNN)代替双三次下采，在有效降低分辨率的同时保留细节信息，得到更为紧凑的低分辨率视频序列，将此低分辨率视频序列通过HEVC帧内编码进行进一步的数据量压缩，通过提出一个质量增强CNN(PPCNN)来改善解码后恢复到原始分辨率的降质视频序列。实验结果显示，本文压缩改进算法在低码率段与标准HEVC相比，能达到更好的质量重建，并且在接近一致的PSNR值时，能节省39.46%的时间和11.04%的比特率，本文算法的视频压缩性能优于HEVC标准算法和相关文献方法。

Improved HEVC intra coding compression algorithm combined with convolutional neural network

In recent years, deep learning of convolutional networks has achieved great success in the fields of image processing and target detection. It has become a new research hotspot to apply Convolutional Neural Networks(CNN) to traditional video compression standards. An improved High Efficiency Video Coding(HEVC) compression algorithm integrated with convolutional neural network is proposed, which integrates down-sampling, HEVC codec, up-sampling and quality enhancement process. In order to extract the structural features of video frames efficiently, two convolutional neural networks are integrated in the proposed compression algorithm. Down Sampling CNN(DwSCNN) replaces bicubic down-sampling, which preserves the detailed information while reducing the resolution, obtaining a more compact low resolution video sequence. The low-resolution video sequence is further compressed by HEVC intra coding, and a quality-enhanced Post Processing CNN(PPCNN) is proposed to improve the degraded video sequence that is restored to the original resolution after decoding. The experimental results show that the proposed compression improvement algorithm can achieve better quality reconstruction than the standard HEVC in the low code rate segment, and can save time by 39.46% and bit rate by 11.04% when the PSNR value is close to the same. The video compression performance of the algorithm is superior to the HEVC standard algorithm and other related literature methods.