Fast Local Tone Mapping, Summed-Area Tables and Mesopic Vision Simulation [chapter]

Marcos Slomp, Michihiro Mikamo, Kazufumi Kane
2012 Computer Graphics  
Introduction High dynamic range (HDR) imaging is becoming an increasingly popular practice in computer graphics, bringing unprecedented levels of realism to computer-generated imagery and rich detail preservation to photographs and films. HDR imagery can embrace more accurately the wide range of light intensities found in real scenes than its counterpart, low dynamic range (LDR), which are tailored to the limited intensities of display devices. Simply put in computer terminology, think of HDR
more » ... a very large, continuous range of intensities encoded in a floating-point representation (but not necessarily), while LDR translates to coarse, quantized ranges, usually encoded as 8-bit integers and thus limited to 256 discrete intensities. An important term in HDR imaging is the dynamic range or contrast ratio of a scene, representing the distance between the lowest and highest intensity values. Luminance in the real world typically covers 14 orders of magnitude (dynamic range of 10 14 : 1), ranging from direct sunlight (10 5 up to 10 8 cd/m 2 ) to shallow starlight (10 −3 down to 10 −6 cd/m 2 ), while typical image formats and commodity display devices can cope with only 2 up to 4 orders of magnitude (maximum contrast ratio of 10 4 : 1) [Ledda et al. (2005) ]. A challenging task on HDR imaging consists in the proper presentation of the large range of intensities within HDR imagery in the much narrower range supported by display devices while still preserving contrastive details. This process involves intelligent luminance (dynamic range) compression techniques, referred to as tone reproduction operators or, more commonly, tone mapping operators (TMO). An example is depicted in Figure 1 . The process of tone mapping shares similarities with the light adaptation mechanism performed by the Human Visual System (HVS), which is also unable to instantly cope with the wide range of luminosity present in the real world. Even though the HVS is only capable of handling a small range of about 4 or 5 orders of magnitude at any given time, it is capable of dynamically and gradually shift the perceptible range up or down, appropriately, in order to better enclose the luminosity range of the observed scene [Ledda et al. (2004) ]. Display devices, on the other hand, are much more restrictive, since there is no way to dynamically improve or alter their inherently fixed dynamic range capabilities. Tone mapping operators can be classified as either global (spatially-uniform) or local (spatially-varying). Global operators process all pixels uniformly with the same parameters, while local operators attempt to find an optimal set of parameters for each pixel individually, often considering a variable-size neighborhood around every pixel; in other words, the 8 www.intechopen.com 2 Will-be-set-by-IN-TECH
doi:10.5772/37288 fatcat:eb73wm3khnarbihmm5jjdt5c3m