Texture denoising needs special training to avoid detail loss
Denoising Textures is a Different Ballgame Altogether!
Computer Vision and Pattern Recognition
Summary
Cleaning up noisy images usually works well, but it often blurs fine textures and patterns. The authors studied how current denoising tools handle textures and found they improve only when specially retrained on texture images. However, this retraining causes the tools to forget what they learned about other kinds of images. This shows that handling textures in image cleanup is a unique challenge that needs dedicated approaches.
What this means in practice
- •For computer vision engineers: Create denoising models that maintain intricate textures by fine-tuning specifically on texture images for better image restoration.
- •For graphics software developers: Improve texture fidelity in creative tools by integrating texture-aware denoisers that avoid blurring detailed patterns.
Authors
Saptarshi Mandal, Oindrila Haldar, Chirayata Bhattacharya, Chandra Sekhar Seelamantula, Zhiwei Xu, Thrasyvoulos N. Pappas, Gaurav Sharma
Abstract
The human brain effortlessly deciphers colors and textures from visual stimuli and reconstructs the world around us, yet preserving such details remains challenging in image denoising. While significant progress has been made in generic image denoising, the problem of texture denoising has remained relatively under-explored. Fine details, such as intricate patterns and repetitive structures, often suffer from over-smoothing even in deep learning based approaches for denoising. In this paper, we analyze the behavior of state-of-the-art (SOTA) denoisers on texture datasets and show that performance improvements are possible only by fine-tuning the denoisers on the texture datasets. Surprisingly, fine-tuning on texture datasets result in the models catastrophically forgetting the original dataset. These findings highlight the need for specialized treatment of textures in denoising. Since denoising quality directly impacts downstream tasks such as generative modelling, we argue that dedicated texture aware denoisers are essential for faithful texture reconstruction and synthesis.