Digital Signal Processing 30 (2014) 74–85 Contents lists available at ScienceDirect Digital Signal Processing www.elsevier.com/locate/dsp Luminance adaptation transform based on brightness functions for LDR image reproduction Hyuk-Ju Kwon, Sung-Hak Lee ∗ , Geun-Young Lee, Kyu-Ik Sohng School of Electrical Engineering and Computer Science, Kyungpook National University, 1370 Sankyug-dong, Buk-gu, Daegu, 702-701, Republic of Korea article info abstract Article history: Available online 1 April 2014 Keywords: Retinex MSR Tone mapping Brightness function Tone mapping algorithms are used for image processing to reduce the dynamic range of an image to be displayed on low dynamic range (LDR) devices. The Retinex, which was developed using multi-scale and luminance-based methods, is one of the tone mapping algorithms for dynamic range compression, color constancy and color rendition. Retinex algorithms still have drawbacks, such as lower contrast and desaturation. This paper proposes a multi-scale luminance adaptation transform (MLAT) based on visual brightness functions for the enhancement of contrast and saturation of rendered images. In addition, the proposed algorithm was used to estimate the minimum and maximum luminance and a visual gamma function for local adapted viewing conditions. MLAT showed enhanced contrast and better color representation than the conventional methods in the objective evaluations (CIEDE200 and VCM).  2014 Elsevier Inc. All rights reserved. 1. Introduction A real world luminance range is much larger than those of dig- ital cameras and display devices. The luminance of blue sky in the morning and shadows in an interior room is approximately 4600 cd/m 2 and less than 10 cd/m 2 , respectively. The differences in these luminance levels can be recognized by human vision. On the other hand, there are limited dynamic range problems when real scenes are captured from a digital camera and displayed on television and monitors. Fig. 1 shows the difference in the visual perception and captured image of a real scene. The digital cam- era needs to adjust its exposure time or shutter speed to capture the scene with a large dynamic range. Such capturing, however, causes saturation in the dark or light areas as well as changes the color appearance of the scenes. To solve these problems, tone map- ping or compression methods for high dynamic range (HDR) image processing have been developed. HDR images are generally rep- resented by different exposure images of the same scene [1] but making an HDR image is difficult when the scenes have a varia- tion in the viewing conditions, such as tilting, shifting and moving objects. Therefore, effective methods that represent an HDR image only using a single LDR image have been developed. Tone mapping algorithms can be divided into two different methods. One is global tone mapping and the other is local tone mapping. Typical global tone mapping algorithms use logarithmic, * Corresponding author. Fax: +82 53 950 5505. E-mail address: shak2@ee.knu.ac.kr (S.-H. Lee). power-law and sigmoid functions. Global mapping algorithms are spatially invariant methods that map the input pixel value to a display value. This can be applied easily to the global contrast en- hancement of the image [2] but it is less effective for images with strong local contrast areas and details. Saturations are shown in the dark or light areas of images and noises are increased in the dark areas. On the other side, local tone mapping algorithms are spatially variant methods. They consider the relationships between the surround pixels and the input pixel value. Therefore, local tone mapping operators can enhance the local contrast and improve de- tailed visibility. Retinex, which is one of the local tone mapping algorithms, was developed by Land [3] as a model of lightness and color perception Fig. 1. Visual perception and captured image reproduction for the real scene. http://dx.doi.org/10.1016/j.dsp.2014.03.008 1051-2004/ 2014 Elsevier Inc. All rights reserved.