Image denoising improved with interpretable neural network blending classical methods
A Study of the Limits of Collaborative DCT-Based Image Denoising via Interpretable Neural Networks
Computer Vision and Pattern Recognition
Summary
Image denoising removes unwanted noise or graininess from pictures, which is important in fields like photography and medical imaging. The authors studied a way to improve traditional denoising methods that rely on clear mathematical steps by making these steps trainable with neural networks. They created a new model called DeepBM3D that mimics a classic method but also learns from data to work better, especially on repeating patterns. Their work shows this approach can match or beat other popular methods while being easier to understand.
What this means in practice
- •For photographers: Reduce noise in photos with a clearer and trainable denoising approach that handles repeating textures effectively.
- •For biomedical imaging teams: Improve image quality in biomedical scans by applying a neural network based denoiser combining classical and learned filtering methods.
Authors
Cristian Comellas, Julia Navarro, Antoni Buades
Abstract
Image denoising remains a fundamental problem in image restoration, with applications in photography, biomedical, and scientific imaging. Modern deep neural networks achieve strong performance by learning powerful image priors, but often rely on large black-box models with limited interpretability. In contrast, DCT-based sliding-window and collaborative filtering methods such as BM3D offer clear algorithmic structure, but depend on handcrafted and non-differentiable operations. This work studies how far such structured collaborative filtering principles can be pushed when reformulated as trainable models. We introduce DeepBM3D, a compact fully differentiable architecture that combines non-local patch grouping, DCT-domain filtering, and multi-stage refinement within a BM3D-inspired pipeline. Lightweight convolutional feature extractors guide patch grouping, while filtering is performed through learned Wiener weights in the DCT domain. Experiments show that DeepBM3D improves over classical and hybrid baselines, remains competitive with FFDNet at low and moderate noise levels, and performs particularly well on repetitive textures.