BLOCK-BASED GRADIENT DOMAIN HIGH DYNAMIC RANGE COMPRESSION DESIGN FOR REAL-TIME APPLICATIONS Tsun-Hsien Wang 1 , Wei-Ming Ke 2 , Ding-Chuang Zwao 2 , Fang-Chu Chen 3 , and Ching-Te Chiu 2 1 SoC Technology Center, Industrial Technology Research Institute, Hsin-Chu, Taiwan, R.O.C. 2 Department of Computer Science, National Tsing Hua University, Hsin-Chu, Taiwan, R.O.C 3 Information & Communications Research Laboratories, Industrial Technology Research Institute, Hsin-Chu, Taiwan, R.O.C. Email: 1 thwang@itri.org.tw, 3 fcchen@itri.org.tw, and 2 ctchiu@cs.nthu.edu.tw ABSTRACT Due to progress in high dynamic range (HDR) capture technologies, the HDR image or video display on conventional LCD devices has become an important topic. Many tone mapping algorithms are proposed for rendering HDR images on conventional displays, but intensive computation time makes them impractical for video applications. In this paper, we present a real-time block-based gradient domain HDR compression for image or video applications. The gradient domain HDR compression is selected as our tone mapping scheme for its ability to compress and preserve details. We divide one HDR image/frame into several equal blocks and process each by the modified gradient domain HDR compression. The gradients of smaller magnitudes are attenuated less in each block to maintain local contrast and thus expose details. By solving the Poisson equation on the attenuated gradient field block by block, we are able to reconstruct a low dynamic range image. A real-time Discrete Sine Transform (DST) architecture is proposed and developed to solve the Poisson equation. Our synthesis results show that our DST Poisson solver can run at 50MHz clock and consume area of 9 mm 2 under TSMC 0.18um technology. Keywords: High Dynamic Range (HDR), gradient, tone mapping, Discrete Sine Transform (DST), Inverse Discrete Sine Transform (IDST), 1. INTRODUCTION The luminance ratio of the natural scenes is about 100,000,000:1 while traditional displays can only show images in dynamic range of 100~1000:1. Due to the technological advances in the HDR capture, high dynamic range images or video become available. Compared with the low dynamic range images, the HDR images show the real scene information much better [2]. However, we are confronted by two challenges. The first is how to display high dynamic range images on the low dynamic range devices, for example, monitors, TV displays, and printers. The second is how to maintain the details of images and to show the complete information in the scene. Tone mapping or tone reproduction is an image processing technique to render the high dynamic range images on conventional displays. Over the past few years, a considerable number of studies have been made on tone mapping. They are commonly classified into global and local tone mappings. Chiu, et al., discover that mapping each pixel of images by global tone operators results in common artifacts because human vision system is nonlinear [3]. They believe that tone mapping based on the feature of pixels results in better effects. However, there is no ideal tone mapping curve that can be applied to every pixel. A method that emphasizes the variety in local areas of images is needed. Because their algorithm is based on local operators, it consumes much computing resource. Furthermore, it is developed based on experimental results rather than theoretical derivations. In 1998, Pattanail, et al., developed a technique to represent patterns, luminance and color processing in human visual systems. They proposed a real scene tone mapping computing model. Their model could process HDR images with perception of scenes at threshold and supra-threshold [4]. Because this algorithm took the adaptability of colors into account, it consumed large amount of computing resource. Nevertheless, it had better sensitivity compared with other tone mapping algorithms. In 2002, Fattal, et al., proposed a simple and effective method to render high dynamic rage images [1]. Their approach manipulated the gradient field of luminance (illumination differences) and attenuated the magnitudes of large gradients. A new low dynamic range image was obtained by solving a Poisson equation on the modified gradient field. It achieved drastic dynamic range compression and well preserved the fine details. Human eyes are more sensitive to the luminance than to colors. Therefore, most HDR compressions process the images on the luminance channel. We adopt the gradient domain high dynamic range compression proposed by Fattal, etc. The reason is that human visual system is more sensitive to illumination differences than to absolute luminance reaching the retina, based on the retinex theory by Land and McCann in 1971[5]. However, the gradient domain HDR compression consumes significant computational time. In [6], it shows that 45.5 sec is required to process a 1600x1200 image on Apple iBook with a G3 processor running at 800MHz. This kind of processing speed cannot meet the requirement for real-time HDR video display. In this paper, we present a block-based gradient domain high dynamic range compression scheme for real-time applications. Instead of processing the whole image at a time, we manipulate the gradient computations block by block. This enhances the processing speed significantly. A fully pipelined fast discrete sine transform (DST) is also presented to solve the Poisson equation. The block-based gradient compression is presented in section II. The hardware implementation and the DST-based Poisson solver are described in section III. The implementation results are shown in section IV. Finally, section V gives a brief conclusion. III - 561 1-4244-1437-7/07/$20.00 ©2007 IEEE ICIP 2007