Robust and interpretable blind image denoising via bias-free convolutional neural networks

TL;DR

This paper introduces a bias-free convolutional neural network architecture for blind image denoising that generalizes better across noise levels and offers interpretability through linear analysis, outperforming bias-dependent models.

Contribution

The authors propose a bias-free CNN architecture that improves robustness and interpretability in blind image denoising, challenging the necessity of biases in deep networks.

Findings

Bias-free networks outperform bias-dependent ones outside training noise levels.

Bias-free architecture maintains state-of-the-art performance within training range.

Linear analysis reveals the network's function as adaptive filtering and subspace projection.

Abstract

Deep convolutional networks often append additive constant ("bias") terms to their convolution operations, enabling a richer repertoire of functional mappings. Biases are also used to facilitate training, by subtracting mean response over batches of training images (a component of "batch normalization"). Recent state-of-the-art blind denoising methods (e.g., DnCNN) seem to require these terms for their success. Here, however, we show that these networks systematically overfit the noise levels for which they are trained: when deployed at noise levels outside the training range, performance degrades dramatically. In contrast, a bias-free architecture -- obtained by removing the constant terms in every layer of the network, including those used for batch normalization-- generalizes robustly across noise levels, while preserving state-of-the-art performance within the training range. Locally, the bias-free network acts linearly on the noisy image, enabling direct analysis of network behavior via standard linear-algebraic tools. These analyses provide interpretations of network functionality in terms of nonlinear adaptive filtering, and projection onto a union of low-dimensional subspaces, connecting the learning-based method to more traditional denoising methodology.