Espoo Vantaa Institute of Technology Mediatekniikan koulutusohjelma Samu Kemppainen Seminar on Media Technology: High Dynamic Range Imaging Seminar Presentation. 4 October 2004 Supervisor: Erkki Rämö, Lecturer
ESPOO VANTAA INSTITUTE OF TECHNOLOGY ABSTRACT Author Name of Thesis Samu Kemppainen High Dynamic Range Imaging Pages Date 13 pages 4 October 2004 Degree Programme Mediatekniikan koulutusohjelma Supervisor Erkki Rämö, Lecturer High dynamic range (HDR) images are used to make more realistic renderings imitating real world lighting in computer. Tone mapping methods is used when displaying HDR images on ordinary display (CRT/LCD). Different encoding methods of saving high dynamic range luminance information are used and the they present differences amounts of visible error. Keywords hdri, dynamic range, tone mapping, radiance maps
3 Contents 1 Introduction 4 2 What is a High Dynamic Range Image? 4 2.1 Acquiring HDR images 4 3 Displaying HDR image on LDR device 6 3.1 Tone mapping 7 3.1.1 Gradient domain HDR compression 7 4 Encoding HDR images 9 5 Some applications of HDR images 12 6 Conclusion 12 References 13
4 1 Introduction The traditional image technologies do not enable us to view a higher luminance rages than 100:1. Human vision has a dynamic range of 1 000 000 000:1, the range from bright sun light to dim star light [1]. In computer generated images, that use global illumination, the range of luminance is much higher that can be displayed on screen. Typically the images displayed on screen are 8-bit per color channel (red/green/blue) witch is 256 intensity levels real-world dynamic range is far greater that that! To store luminance information in the high dynamic range images we have to use different techniques than in traditional image. Different methods of encoding is used to encompass a large range of values in to limited rage of bits. 2 What is a High Dynamic Range Image? HDR image is a image that holds higher dynamic range that can be show on traditional display device (CRT/LCD) or that can be captured with a camera with just a single exposure [2]. The "dynamic range" of a scene is the contrast ratio (also know as luminance range) between its brightest and darkest parts. HDR image is useful for representing true illumination values in image-based rendering applications. They are also useful for recording incident illumination and using it to illumination CG objects for realistic composition. 2.1 Acquiring HDR images Acquiring images with high dynamic range can be done in various ways: computing them on computer or collect the luminance data from a real-word environment. One way of rendering HDR image is called Physically Based Spectral rendering [3] which uses group of image synthesis methods that are representations of the associated light
5 spectrum of color values. These methods have their advantages but the problem is that they are too complicated. Normal digital (or analog) camera can be used to collect real-world luminance information by taking series of images with different exposure times (Figure 1) at the same position. This series is then combined into a single high dynamic range image at HDR Shop [2] or similar program. Figure 1: 16 differently exposed photographs from 30 seconds to 1/1000 seconds [4]. Recovering High Dynamic Range Radiance Maps from Photographs is a technique that uses this multi exposure method of collecting high luminance range. High dynamic range images are useful for representing true illumination values in image-based rendering applications. In 3-dimensional rendering software HDR images can be used to illuminate CG object or environment for realistic composition. It is possible to use an flat image as a radiance map but when high dynamic scene is photographed using a light probes (a mirror ball). Other techniques involve stitching multiple images together or using scanning panoramic camera. [4] The advantages of using HDR images
6 Figure 2: a) Original photograph, b) Motion-blurred LDR image, c) Motion-blurred HDR image and d) Photograph with real motion blur. 3 Displaying HDR image on LDR device To view high dynamic range images on low dynamic range device (for example CRT/LCD) it is possible to use pure software solution, limited resources of a 3Dhardware or hardware that support floating point color spaces (a.k.a. HDR images). State-of-the-art hardware is capable of processing a real time images in HDR mode and convert them in LDR just at the end of pipeline. When using Real Time High Dynamic Range Texture Mapping [6] textures are decomposed into sets of 8-bit textures. Then they are dynamically reassembled by the graphics hardware's multi texturing system or by using multi pass techniques and frame buffer image processing. This technique needs only two 8-bit textures to be in the memory simultaneously. (See figure 3.) With this technique is it possible to use 8-bit textures to be used to create hardware accelerated real time HDR graphics. [6] Rounding errors occur in HDR rendering pipeline when multiple passes on framebuffer information is done. This will degrade the image quality (on 24-bit image) and cause artifacts on image. Even on 64-bit floating point images can have notable amounts of errors comparing to 128-bit images. Visual quality can be maintained by using all the data in high precision right until it is going to be shown on LDR device. Matter about rounding error is explained on section 4: Ending HDR images
7 Figure 3: Real time HDR texture mapping pipeline [6] 3.1 Tone mapping Tone mapping takes advantages of Human Visual System (HVS), the fact that HVS has a greater sensitivity to relative luminance levels rather than absolute. Tone mapping is what brings the high dynamic range image to display device. Tone Reproduction and Physically Based Spectral Rendering [3] is developed for use in television and photography. 3.1.1 Gradient domain HDR compression Gradient Domain HDR compression is a way of rendering HDR images on conventional display. It is very fast and simple to use. it method is used to manipulate
8 the gradient field of the luminance image by attenuating the magnitudes of large gradients. Then a new, low dynamic range image is obtained by solving a Poisson equation on the modified gradient field. [7] Figure 4: a) Gradient attenuation factors used to compress HDR radiance map b) the result [7] Gradient domain HDR compression is also useful in ordinary photography. By using only two differently exposed images it is possible to combine a one finely balanced image. (See figure 5.)
9 Figure 5: Example of gradient domain HDR compression in ordinary photographs [7] 4 Encoding HDR images High dynamic range images have to be encoded to encompass a large range of values. Most used encoding methods are listed in table 1. Table 1: Most used encodings [8] Pixar Log Encoding Radiance RGBE Encoding SGI LogLuv ILM OpenEXR Microsoft/HP scrgb Encoding (TIFF) (HDR) (TIFF) (EXR)
10 Most of the encoding methods use logarithmic encoding of luminance values to compress the color information to certain amount of bits. Benefits of log and floating point representations over linear or gamma encodings: - 24 bits LogLuv format holds more dynamic range than the 36-bit scrgb-nl format (gamma encoding) and even the 48-bit scrgb linear encoding - 32 bits LogLuv encoding holds 10 times the dynamic range over the scrgb and scrgb-nl. - The EXR encoding holds 3 times the range of scrgb encoding in the same 48 bits, with much higher precision than any of the other formats Figure 6: Cost (bit/pixel) versus benefit (dynamic range) of full gamut formats [8] Some rounding errors occur when using log or floating point formats but when using enough bits the error levels are quite low. Example of the rounding errors can be seen in figure 7. The quality of different encodings can be seen on figure 8.
11 Figure 7: False color difference view of 24-bit LogLuv encoding [8] Figure 8: Encoding quality curves (average) [8]
12 5 Some applications of HDR images High dynamic range imaging have wide variety of usage including global illumination in physically-based renderings, mixed reality renderings (special effects for movies and commercials), human vision simulation and psychophysics, reconnaissance and satellite imaging (remote sensing), digital composing for film, digital cinema etc. 6 Conclusion With High dynamic range (HDR) images it is possible to make more realistic renderings imitating real world lighting in computer. Using Tone mapping methods when displaying HDR images enables use to see wider dynamic range compressed to ordinary display (CRT/LCD) luminance range. Different encoding methods of saving high dynamic range luminance information are used preserve the disk space although they present some amounts of visible error kind of seen in jpeg compression.
13 References 1 Devlin Kate. A review of tone reproduction techniques. 2002 2 HDR Show Home - Introduction. WWW-ducument. http://www.ict.usc.edu/graphics/hdrshop. 1 October 2004. 3 Devlin, Chalmers, Wilkie and Purgatholfer. Tone Reproduction and Physically Based Spectral Rendering. 2002 4 Debevec and Malik. Recovering High Dynamic Range Radiance Maps from Photographs. 1997. 5 Light Progbe Image Gallery. WWW-document. http://www.debevec.org/probes/. 1 October 2004. 6 Cohen, Tehou, Hawkins and Debevec. Real Time High Dynamic Range Texture Mapping. 2001. 7 Fattal, Linschinski and Werman. Gradient Domain High Dynamic Range Compression. 2002 8 High Dymanic Range Image Encodings. WWW-document. http://www.anyhere.com/gward/hdrenc/. 1 October 2004.
14 Streamit Kommentteja viidestä esityksestä joissa en ollut itse paikalla. Olen jokaisesta pyrkinyt kommentoimaan esiintymistä ja puheen selkeyttä, kalvoja ja joitain hyvä ja huonoja asioita. Reija Mehtiö: Design for all Reijan esiintyminen on rauhallista ja puheesta saa selvää. Välillä joidenkin sanojen ääntäminen on hieman epäselvä, mutta kun tietää asian mistä hän puhuu, niin ymmärtää niiden merkityksen. Kalvot ovat selkeät. Niissä ei ole liikaa tekstiä, joskus jopa hieman liian vähän kuten guidelines kalvossa, jossa olisi toivonut olevan edes hieman selitystä. Pekka Kinnunen: Mobile games Pekan puhe on selkeätä ja sujuvaa. Välillä hän liikehtii edessä hieman häiritsevästi edes takaisin, mutta muuten esiintyminen on hyvää. Kalvot ovat selkeitä ja eivät sisältäneet liikaa tekstiä. Ajankäytöllisesti esitys jäi hieman liian lyhyeksi. Nima Ehsani: Hackers Niman esiintyminen oli erittäin rauhatonta, hän liikehti edes takaisin. Historia osuus oli mielenkiintoinen ja huomasi kyllä, että Nima oli perehtynyt aiheeseen kohtuullisen hyvin. Hän puhuu hakkeri/krakkeri termien käytöstä, mutta käyttää niitä paikoin kuitenkin itse hieman sekavasti. Kalvoilla oli liikaa tekstiä ja välillä Nima luki suoraan niistä. (Powerpoint-tiedostoa ei saatavilla.)
15 Minna Solismaa: Digital TV - advertising Minnan esiintyminen on erittäin rauhallista ja puhe selkeää. Välillä tosin tuntui ettei hän pääse oikein eteenpäin asiassa. Kalvot olivat selkeät ja niissä oli tarpeeksi runkoa tukemaan Minnan puhetta. Aiheen valinta oli mielenkiintoinen. (Powerpoint-tiedostoa ei saatavilla.) Liisa Benmergui: Filesharing and the birth of the digital music industry Liisan esiintyminen on rauhallista ja puhe erittäin selkeää. Kalvoilla oli välillä paljon tekstiä, mutta suurin osa oli kyllä tarpeellista eli se selkeytti asian (lukujen jne.) ymmärtämistä. Kuvat olivat selkeitä ja helposti ymmärrettäviä.