2020/03/05 by Tae‐Eon Park, Taesup Moon, Park, Tae-Eon +1 · 1 citation
Computer Science · #Blind Source Separation Techniques #Digital Filter Design and Implementation #FOS: Computer and information sciences #Image and Signal Denoising Methods #Information Theory (cs.IT) #Machine Learning (cs.LG) #Machine Learning (stat.ML)
paper · pdf · doi:10.48550/arxiv.2003.02623
openalex publication_date 2020/03/05 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
We devise a novel neural network-based universal denoiser for the\nfinite-input, general-output (FIGO) channel. Based on the assumption of known\nnoisy channel densities, which is realistic in many practical scenarios, we\ntrain the network such that it can denoise as well as the best sliding window\ndenoiser for any given underlying clean source data. Our algorithm, dubbed as\nGeneralized CUDE (Gen-CUDE), enjoys several desirable properties; it can be\ntrained in an unsupervised manner (solely based on the noisy observation data),\nhas much smaller computational complexity compared to the previously developed\nuniversal denoiser for the same setting, and has much tighter upper bound on\nthe denoising performance, which is obtained by a theoretical analysis. In our\nexperiments, we show such tighter upper bound is also realized in practice by\nshowing that Gen-CUDE achieves much better denoising results compared to other\nstrong baselines for both synthetic and real underlying clean sequences.\n