Blind compression artifact reduction using dense parallel convolutional neural network
作者：
Highlights：
• Parallel convolutional layer provides model parallelism and achieves faster training.
• Efficient flow of information and gradients is achieved with dense skip connections.
• Reconstruction accuracy is improved with collective knowledge of the feature representations.
• Bottleneck layer reduces the computational complexity and improves the model efficiency.
摘要
•Parallel convolutional layer provides model parallelism and achieves faster training.•Efficient flow of information and gradients is achieved with dense skip connections.•Reconstruction accuracy is improved with collective knowledge of the feature representations.•Bottleneck layer reduces the computational complexity and improves the model efficiency.
论文关键词：Compression artifacts,Feature redundancy,Lossy compression,Parallel convolution,Quality factor,Skip connection
论文评审过程：Received 2 December 2019, Revised 6 August 2020, Accepted 9 September 2020, Available online 14 September 2020, Version of Record 24 September 2020.
论文官网地址：https://doi.org/10.1016/j.image.2020.116009
