Preventing Overfitting in Deep Image Prior for Hyperspectral Image Denoising
2026-04-09 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors study a way to improve a deep learning method called Deep Image Prior (DIP) used for cleaning noisy hyperspectral images. Normally, DIP tends to overfit, which means it starts to memorize the noise instead of removing it properly, so it needs early stopping. Their new method reduces overfitting by combining a special loss that is less sensitive to outliers with a technique that controls the model's sensitivity during training. Tests on real noisy images show their approach works better than previous DIP methods for this task.
Deep Image PriorHyperspectral Image DenoisingOverfittingSmooth L1 LossRegularizationData FidelityInverse Imaging ProblemsGaussian NoiseSparse NoiseStripe Noise
Authors
Panagiotis Gkotsis, Athanasios A. Rontogiannis
Abstract
Deep image prior (DIP) is an unsupervised deep learning framework that has been successfully applied to a variety of inverse imaging problems. However, DIP-based methods are inherently prone to overfitting, which leads to performance degradation and necessitates early stopping. In this paper, we propose a method to mitigate overfitting in DIP-based hyperspectral image (HSI) denoising by jointly combining robust data fidelity and explicit sensitivity regularization. The proposed approach employs a Smooth $\ell_1$ data term together with a divergence-based regularization and input optimization during training. Experimental results on real HSIs corrupted by Gaussian, sparse, and stripe noise demonstrate that the proposed method effectively prevents overfitting and achieves superior denoising performance compared to state-of-the-art DIP-based HSI denoising methods.