A Retinex-Based Local Tone Mapping Algorithm Using L0 Smoothing Filter - Advances in Image and Graphics Technologies - page 40

A Retinex-Based Local Tone Mapping Algorithm
Using L 0 Smoothing Filter

Lei Tan, Xiaolin Liu * , and Kaichuang Xue

College of Mechatronic Engineering and Automation
National University of Defense Technology
Changsha, Hunan, 410073, P.R. China
{tanlei08,lxlchangsha,xkcwork}@sina.com

Abstract. In this paper, we propose a novel halo-free local tone mapping
algorithm using L 0 smoothing filter. Our method imitates the adaptation of the
mechanism of the human visual system (HVS), which ensures a strong
adaptability to the scenes of different dynamic ranges. Firstly, we will apply a
global histogram adjustment method to the luminance image, which is a simple
initial global adaptation; secondely, we will demonstrate how the luminance
image is remapped by a retinex-based local tone mapping method. During the
estimation of illumination, a L 0 smoothing filter is used instead of Gaussian filter
to compress the contrast while reducing the halo artifacts; and finally, through the
color correction, we will show how the tone-mapped RGB image is obtained.
According to our experimental results, the proposed method outperforms the
state-of-art tone mapping algorithms in color rendition and detail preservation.

Keywords: High dynamic range, tone mapping, retinex, L 0 smoothing filter,
color correction.

1 Introduction

Dynamic range is defined by the luminance ratio of the highest scene luminance to
the lowest, that is, the number of levels is divided in an image from the darkest
grayscale to the brightest. An image is said to be at a High Dynamic Range (HDR)
when its dynamic range exceeds by far of the display device. It's understood that an
image is at the Low Dynamic Range (LDR) when its dynamic range is below the
display device. Tone mapping algorithm is a method that compresses the dynamic
range of the HDR images so that the mapped image can fit into the dynamic range of
the display devices.
There are many tone mapping algorithms which can be divided into two categories:
global and local tone mapping operators. The global tone mapping operators are
computationally very simple since all the pixels of the original image are mapped
using the same transformation. But one should be aware that the global tone mapping
operators may cause the loss of details in the dark and bright areas of HDR image
when the dynamic range of the image is too large. The local tone mapping operators
* Corresponding author.