DPW-SDNet: Dual Pixel-Wavelet Domain Deep CNNs for Soft Decoding of JPEG-Compressed Images

JPEG is one of the widely used lossy compression methods. JPEG-compressedimages usually suffer from compression artifacts including blocking andblurring, especially at low bit-rates. Soft decoding is an effective solutionto improve the quality of compressed images without changing codec orintroducing extra coding bits. Inspired by the excellent performance of thedeep convolutional neural networks (CNNs) on both low-level and high-levelcomputer vision problems, we develop a dual pixel-wavelet domain deepCNNs-based soft decoding network for JPEG-compressed images, namely DPW-SDNet.The pixel domain deep network takes the four downsampled versions of thecompressed image to form a 4-channel input and outputs a pixel domainprediction, while the wavelet domain deep network uses the 1-level discretewavelet transformation (DWT) coefficients to form a 4-channel input to producea DWT domain prediction. The pixel domain and wavelet domain estimates arecombined to generate the final soft decoded result. Experimental resultsdemonstrate the superiority of the proposed DPW-SDNet over severalstate-of-the-art compression artifacts reduction algorithms.