Deep Image Prior using Stein's Unbiased Risk Estimator: SURE-DIP

TL;DR

This paper introduces SURE-DIP, a novel method combining Stein's unbiased risk estimate with deep image prior to reduce overfitting in single-shot image recovery, improving performance without extensive training data.

Contribution

The paper proposes a generalized SURE loss for DIP, effectively minimizing overfitting and enhancing image recovery performance, including with unrolling architectures.

Findings

SURE-DIP reduces overfitting compared to classical DIP.

SURE-DIP outperforms traditional DIP in image recovery tasks.

Integration with unrolling architectures further improves results.

Abstract

Deep learning algorithms that rely on extensive training data are revolutionizing image recovery from ill-posed measurements. Training data is scarce in many imaging applications, including ultra-high-resolution imaging. The deep image prior (DIP) algorithm was introduced for single-shot image recovery, completely eliminating the need for training data. A challenge with this scheme is the need for early stopping to minimize the overfitting of the CNN parameters to the noise in the measurements. We introduce a generalized Stein's unbiased risk estimate (GSURE) loss metric to minimize the overfitting. Our experiments show that the SURE-DIP approach minimizes the overfitting issues, thus offering significantly improved performance over classical DIP schemes. We also use the SURE-DIP approach with model-based unrolling architectures, which offers improved performance over direct inversion schemes.