2017/03/29 by Nick Johnston, Damien Vincent, Johnston, Nick +15 · 11 citations
Computer Science · #Advanced Data Compression Techniques #Advanced Image Processing Techniques #Algorithm #Artificial intelligence #Codec #Computer Vision and Pattern Recognition (cs.CV) #Computer science #Computer vision #Convolutional neural network #Data compression #ENCODE #Entropy encoding #FOS: Computer and information sciences #Image (mathematics) #Image and Signal Denoising Methods #Image compression #Image processing #Image quality #JPEG #JPEG 2000 #Lossless compression #Lossy compression #Pattern recognition (psychology) #Quantization (signal processing) #cs.CV
paper · pdf · doi:10.48550/arxiv.1703.10114
published in arXiv (Cornell University) (Cornell University)
arxiv created 2017/03/29 · openalex publication_date 2017/03/29 · arxiv updated 2017/03/30 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/08
We propose a method for lossy image compression based on recurrent,\nconvolutional neural networks that outperforms BPG (4:2:0 ), WebP, JPEG2000,\nand JPEG as measured by MS-SSIM. We introduce three improvements over previous\nresearch that lead to this state-of-the-art result. First, we show that\ntraining with a pixel-wise loss weighted by SSIM increases reconstruction\nquality according to several metrics. Second, we modify the recurrent\narchitecture to improve spatial diffusion, which allows the network to more\neffectively capture and propagate image information through the network's\nhidden state. Finally, in addition to lossless entropy coding, we use a\nspatially adaptive bit allocation algorithm to more efficiently use the limited\nnumber of bits to encode visually complex image regions. We evaluate our method\non the Kodak and Tecnick image sets and compare against standard codecs as well\nrecently published methods based on deep neural networks.\n