Zheng Bolun ; Chen Yaowu ; Tian Xiang ; Zhou Fan ; Liu Xuesong

Abstract

Several dual-domain convolutional neural network-based methods showoutstanding performance in reducing image compression artifacts. However, theysuffer from handling color images because the compression processes forgray-scale and color images are completely different. Moreover, these methodstrain a specific model for each compression quality and require multiple modelsto achieve different compression qualities. To address these problems, weproposed an implicit dual-domain convolutional network (IDCN) with the pixelposition labeling map and the quantization tables as inputs. Specifically, weproposed an extractor-corrector framework-based dual-domain correction unit(DCU) as the basic component to formulate the IDCN. A dense block wasintroduced to improve the performance of extractor in DRU. The implicitdual-domain translation allows the IDCN to handle color images with thediscrete cosine transform (DCT)-domain priors. A flexible version of IDCN(IDCN-f) was developed to handle a wide range of compression qualities.Experiments for both objective and subjective evaluations on benchmark datasetsshow that IDCN is superior to the state-of-the-art methods and IDCN-f exhibitsexcellent abilities to handle a wide range of compression qualities with littleperformance sacrifice and demonstrates great potential for practicalapplications.