BRIGHTNESS PRESERVING VIDEO CONTRAST ENHANCEMENT USING S-SHAPED TRANSFER FUNCTION Ke Gu, Guangtao Zhai, Min Liu, Qi Xu, Xiaokang Yang, and Wenjun Zhang Institute of Image Communication and Information Processing, Shanghai Jiao Tong University, Shanghai, China Shanghai Key Laboratory of Digital Media Processing and Transmissions ABSTRACT This paper presents an efficient perceptual model inspired efficient video contrast enhancement algorithm. We propose a S-shaped transfer function for image pixel values that ef- fectively improves the perceived contrast while preserving brightness of the scene. The S-shaped transfer function has only one control parameter that can be adaptively chosen for different video contents, such as sports, cartoon, news, and landscape programs. Then, the input image brightness is further preserved, in order to maintain the perception of human visual system (HVS) to some special scenes, such as dark scene and seaside scene. Experiments and compara- tive study on VQEG Phase I test database demonstrate that the proposed S-shaped Transfer function based Brightness Preserving (STBP) contrast enhancement algorithm outper- forms various histogram equalization based methods such as HE, DSIHE, RSIHE and WTHE, yet with much lower computational complexity. Index Terms— Video contrast enhancement, S-shaped transfer curves, human visual system (HVS), video scenes, brightness preservation 1. INTRODUCTION Contrast enhancement is an important research topic for im- age processing and computer vision. By smartly redistribut- ing pixel values in an image, the image contrast can be dras- tically improved, as practiced by the traditional histogram e- qualization (HE) [1]. The fundamental objective of HE is to achieve the maximum entropy of the processed image so as to preserve image details. In brief, HE is conducted by adjusting pixel values according to the probability distribution of the input image pixel. Nowadays, the classical HE method has been widely employed in many image/video post-processing systems, because of its simplicity and effectiveness. Howev- er, since HE does not preserve the mean brightness and can therefore sometimes cause visible deterioration. More and more researchers in image processing tend to agree that HE is far from ideal as an image contrast enhancement algorithm. Realizing this major drawback of HE, a large number of improved approaches by direct modification of HE to pre- serve the brightness have been proposed. Early methods BB- HE [2] and DSIHE [3] by first decompose the input image histogram into dualistic sub-histograms, and then indepen- dently perform HE in each sub-histogram. The distinction is that the decomposition step of BBHE relies on mean bright- ness, while DSIHE using median brightness. To better pre- serve the mean/median brightness during sub-histogram sep- aration, RMSHE [4] and RSIHE [5] adopt recursive oper- ation to improve BBHE and DSIHE methods, respectively. In addition, some contrast enhancement techniques of high- er computational load were recently developed based on the manipulation of dynamic range [6, 7, 8, 9]. Despite of the abundance of image contrast enhancement techniques, only very few studies have been devoted to video contrast enhancement during the last decade. Existing meth- ods for video contrast enhancement can be divided into two categories. The first type of algorithms are the direct exten- sion of image based methods. For instance, DSIHE, RSIHE and other brightness preserving algorithms can be applied to each frame. The WTHE algorithm [10] modifies image his- togram by weighting and thresholding followed by HE. In the second type, [11] proposed a new and robust video contrast enhancement approach, by analyzing video streams and clus- ter frames that are similar to each other. More specifically, they extract key frames belonging to each cluster using eigen analysis and estimate enhancement parameters for only the key frame, and then use these parameters to enhance frames belonging to that cluster. It is easy to imagine that video contrast enhancement de- mands lower computational complexity and higher temporal consistency. Those requirements rule out complicated method such as [6, 7, 8, 9, 11] as well as HE-based algorithms such as [1, 2, 3, 4, 5, 10] that may cause temporal luminance fluctua- tion due to the possible combination of neighboring histogram bins or the lack of brightness preservation. Our recent study indicates that a simple S-shaped transfer function is quite ef- fective in perceptual contrast enhancement. The transfer func- tion is controlled by only one free parameter that is tunable to different visual contents to customize and further improve the performance. Moreover, the S-shaped transfer function pre- serves the median brightness so it can be safely used in video sequence without spoiling the perception of human visual sys-