A Novel Variational Model for Detail-Preserving Low-Illumination Image Enhancement

Yadong Xu,Beibei Sun

DOI: https://doi.org/10.1016/j.sigpro.2022.108468

IF: 4.729

2022-06-01

Signal Processing

Abstract:The images captured under low-light conditions often show low sharpness and low contrast, which inevitably influences the quality of the images and degrades the performance of many computer vision systems designed for pattern recognition. Although many low-light image enhancement methods have been put forward to deal with the problem, the existing methods may bring about contrast distortion and insufficient or excessive enhancement of the source images. To tackle this challenge, a novel variational enhancement model is presented. This method focuses on enhancing the visibility of low-illumination images while preserving the details and the texture information. To begin with, we introduce a new variational model, which includes novel regularization terms for the illumination, the reflectance and the noise component. Specifically, we use the l 1 norm to implement the piece-wise continuousness operation on the reflectance map, present a novel structure-preserving regularizer to assure the scale-aware structure smoothness of the illumination map, and take noise map estimation into account to suppress the noise of the given image. Then, an alternating direction method of multipliers (ADMM) is utilized to solve the specified problem accurately. The extensive experimental results have shown that the proposed approach yields comparative and even better performance in comparison with some state-of-the-art techniques in both qualitative and quantitative evaluations.

engineering, electrical & electronic