Deep Image Prior with L0 Gradient Regularizer for Image Smoothing

TL;DR

This paper introduces DIP-ℓ0, a deep image prior framework with an ℓ0 gradient regularizer that achieves high-quality, training-data-free image smoothing, outperforming existing methods in edge preservation and artifact removal.

Contribution

It proposes a novel DIP framework incorporating an ℓ0 gradient regularizer and an ADMM-based optimization algorithm, enabling effective training-data-free image smoothing.

Findings

Outperforms many existing image smoothing algorithms

Effective in edge-preserving smoothing and JPEG artifact removal

Demonstrates high-quality results without training data

Abstract

Image smoothing is a fundamental image processing operation that preserves the underlying structure, such as strong edges and contours, and removes minor details and textures in an image. Many image smoothing algorithms rely on computing local window statistics or solving an optimization problem. Recent state-of-the-art methods leverage deep learning, but they require a carefully curated training dataset. Because constructing a proper training dataset for image smoothing is challenging, we propose DIP-$\ell_0$, a deep image prior framework that incorporates the $\ell_0$ gradient regularizer. This framework can perform high-quality image smoothing without any training data. To properly minimize the associated loss function that has the nonconvex, nonsmooth $\ell_0$ ``norm", we develop an alternating direction method of multipliers algorithm that utilizes an off-the-shelf $\ell_0$ gradient minimization solver. Numerical experiments demonstrate that the proposed DIP-$\ell_0$ outperforms many image smoothing algorithms in edge-preserving image smoothing and JPEG artifact removal.