基于正则化高斯场模型的低光图像增强

利用交替方向优化技术的Retinex分解是一种有效的低光图像增强解决思路。但是由于反射层和照度层的联立估计通常被视为一个多块凸优化问题,这导致算法结构复杂,优化效率低。本文将Retinex分解扩展到高斯场中,提出了一种基于正则化高斯场(Regularized Gaussian Fields,RGF)模型的低光图像增强方法。通过建立基于RGF的目标函数,将反射层和照度层的联立估计表述为一个无约束优化问题。该目标函数是可微的,因此可以通过标准的梯度优化技术进行解优化,同时解出反射层和照度层。然后,利用高斯核权重对求解出的反射层进行校正,以避免过度曝光而导致的细节丢失。经过大量的定性和定量对比实验,结果说明了与目前公认效果较好的方法相比,本文方法在增强效率和质量方面都具有一定优势。

Low light enhancement based on regularized Gaussian fields model

Retinex decomposition with alternating direction minimization techniques is a prevalent solution for low light enhancement. This is because estimating the reflectance and illumination is typically considered as a multi block convex optimization problem,resulting in complex algorithm structure and low optimization efficiency. In this paper,we propose a regularized Gaussian fields(RGF)model for low light enhancement. The RGF based optimization function is established to formulate simultaneous reflectance and illumination estimation as an unconstrained optimization problem. This function is differentiable and hence could be minimized by standard gradient based optimization techniques. Then,the reflectance solved from the RGF based optimization function is refined based on the Gaussian kernel to avoid over exposure and preserve naturalness. The qualitative and quantitative comparisons on challenging low light images demonstrate the superiority of our method over several state of the arts in terms of enhancement efficiency and quality.