🤖 AI Summary
Microscopy images are commonly corrupted by compound noise combining signal-dependent and row/column-correlated components, which existing self- or unsupervised denoising methods struggle to jointly model. To address this, we propose the first fully unsupervised deep learning denoising framework—requiring neither paired data, predefined noise models, nor clean reference images. Our method employs a dual-decoder variational autoencoder (VAE), where an autoregressive decoder explicitly captures spatial dependencies among pixels, while latent-space constraints enforce encoding of only the underlying clean signal. Evaluated on multimodal microscopy datasets, our approach significantly outperforms state-of-the-art self- and unsupervised denoisers. It is the first to achieve end-to-end separation of compound noise comprising both row/column correlations and signal dependence. Moreover, it demonstrates strong generalization under zero-shot inference, without any task-specific training.
📝 Abstract
Accurate analysis of microscopy images is hindered by the presence of noise. This noise is usually signal-dependent and often additionally correlated along rows or columns of pixels. Current self- and unsupervised denoisers can address signal-dependent noise, but none can reliably remove noise that is also row- or column-correlated. Here, we present the first fully unsupervised deep learning-based denoiser capable of handling imaging noise that is row-correlated as well as signal-dependent. Our approach uses a Variational Autoencoder (VAE) with a specially designed autoregressive decoder. This decoder is capable of modeling row-correlated and signal-dependent noise but is incapable of independently modeling underlying clean signal. The VAE therefore produces latent variables containing only clean signal information, and these are mapped back into image space using a proposed second decoder network. Our method does not require a pre-trained noise model and can be trained from scratch using unpaired noisy data. We benchmark our approach on microscopy datatsets from a range of imaging modalities and sensor types, each with row- or column-correlated, signal-dependent noise, and show that it outperforms existing self- and unsupervised denoisers.
