Abstract
Recently, DNN models for lossless image coding have surpassed theirtraditional counterparts in compression performance, reducing the bit rate byabout ten percent for natural color images. But even with these advances,mathematically lossless image compression (MLLIC) ratios for natural imagesstill fall short of the bandwidth and cost-effectiveness requirements of mostpractical imaging and vision systems at present and beyond. To break thebottleneck of MLLIC in compression performance, we question the necessity ofMLLIC, as almost all digital sensors inherently introduce acquisition noises,making mathematically lossless compression counterproductive. Therefore, incontrast to MLLIC, we propose a new paradigm of joint denoising and compressioncalled functionally lossless image compression (FLLIC), which performs losslesscompression of optimally denoised images (the optimality may be task-specific).Although not literally lossless with respect to the noisy input, FLLIC aims toachieve the best possible reconstruction of the latent noise-free originalimage. Extensive experiments show that FLLIC achieves state-of-the-artperformance in joint denoising and compression of noisy images and does so at alower computational cost.