Halo-free image enhancement through multi-scale detail sharpening and single-scale contrast stretching

Abstract

Conventional image enhancement methods manipulate images in just one scale. This is usually not enough in practice since details like edges live across scales. So, enhancing images through multiscale manipulations is necessary to achieve satisfying results. One drawback of conventional multiscale enhancement methods is that halo artifacts are easy to be introduced. In this paper, we propose a novel efficient multi-scale halo-free image enhancement method, which is based on the observation that image enhancement can be decomposed into two layers: detail sharpening and contrast stretching, and multiscale operations need to be employed to process just the detail layer. We then propose to decompose the multiscale detail manipulations into one-dimensional edge-preserving operations such that halos are completely avoided. For the contrast stretching layer, a modified histogram equalization method is proposed, which achieves state-of-the-art global contrast enhancing effect. Numerical experiments show that, compared to existing popular approaches, the proposed method produces quite competitive results for a variety of image types.