CSE 564: Visualization Color - Cs.stonybrook.edu

Perception of Light IntensityCSE 564: VisualizationColorKlaus MuellerComputer Science DepartmentStony Brook UniversityHow Many Intensity Levels Do We Need?Dynamic Intensity Range IssuesDynamic range of the natural world:100 000 000:1Dynamic range the eye can accommodate in a single view:10 000:1Dynamic range a typical monitor can display:100:1Dynamic range a typical camera can capture:100:1

Long Camera ExposureShort Camera Exposurecaptured the interior well, but the outside is too bright captured the outside well, but now the interior is dark Medium Camera ExposureWhat Now OK. now we have three images that have captured all thedetail of the scene but we want to visualize it all in one picture, not three we need some way to merge these three pictures this is the domain of High Dynamic Range Imaging (HDR)How does HDR work?Two methods: somehow compress the large range into a small, displayable range look at small neighborhoods and try to maximize contrast in each the first is a global method, the second is a local methodThis is also often called tone mappingeverything is somewhat present, but not very detailedAnother application of HDR: computational datasets are often computed in floating point precision HDR can be used to compress the floating point images into 8-bit

MethodsGlobal methods: scale each pixel according to a fixed curve the key issue is here: the shape of the curveMethodsLocal methods: group small neighborhoods by their average value scale these averages down add detail back inlocalsmoothing Local methods: group small neighborhoods by their average value scale these averages down add detail back inbeforescale downadd detailback inafterComparison: Global MethodComparison: Local Method

Result With Earlier ExampleExample: Grand CanalExample: Grand CanalExample: Grand Canal

Example: Grand CanalExample: FoyerExample: FoyerExample: Foyer

Example: FoyerReferencesHDR has become a popular techniqueSome of the key HDR researchers are: P. Debevec, E. Reinhard, G. Ward, M. Ashikhmin, J. Tumblin, andothers for use of HDR in scientific visualization, see X. Yuan, M. Nguyen, B.Chen and D. Porter, “High Dynamic Range Volume Visualization,”IEEE Transaction on Visualization and Computer Graphics, vol. 12,no. 4, 2006.Image examples were taken from http://www.hdrsoft.comBack to The Optical Illusion ExampleExplanationWhile the retina can perceive a high range of intensities, itcannot handle all simultaneously at any given time, each region adapts to a small intensity rangedetermined by the local intensity that is why you have to wait a while when you step from a bright intoa dark room (say, a dark movie theater from a brightly lit lobby)after moving the eye:eventually adaptedrangecurrent adaptedrangeeventually the bright areaintensity is unsaturated,matches neighborhood(which was already adaptedhere before)after moving the eye:new bright area saturatesintensity perceptioncurrent dark areain picture falls here

Spectrum of WavelengthsPerception Curvescolor generation with primarieshuman color sensitivity curvesPerceptional Color SpacesUse Of The CIE Chromaticity Diagram

The Munsell Perceptional Color SpaceThe (irregularly shaped) Munsell tree has 3 axes: chroma (saturation): distance from the core (values 0-30, with fluorescent colors having the maximum 30)value (brightness): vertical axis (0– 10 (white))hue: 10 principal hues (R, YR, Y, GY, G, BG,B, PB, P, RP)Non-Perceptional Color are to: CIE LAB in 3DApplication: Colorization of Grey-Level ImagesApplication: Colorization of Grey-Level Imagesmovie:

Application: Colorization of Grey-Level ImagesReferencesMore information: T. Welsh, M. Ashikhmin, and K. Mueller, "Transferring color togreyscale images," ACM Transactions on Graphics (Proc. ofSIGGRAPH'02), vol. 21, no. 3, pp. 277-280, 2002.movie:More on ColorLabeling vs. Contrastfrom: M. Stone

More on ColorMore on ColorUse of ColorLuminance Contrast

Luminance ContrastColor Contrast and HarmonyColor HarmonyHarmonic Color SchemesNon-harmonic colorsHue wheel:Harmonic colorsi typeV typeL typeI typeT typeY typeX typeN type

Color Harmonization Procedure (1)Color Harmonization Procedure (2)Given arbitrary hue histogram H(p) for image X, find theclosest harmonic template Tm minimize the distance of the histogram to template coverage(delineated by template edges E)width w use an optimization procedure for this also find the orientation angle αF ( X ( m, α )) H ( p) Ep XGiven closest template and α has been found (user mayspecify other template) shift all hues H(p) to the closest harmonic template position H’(p) withTm (α ) a Gaussian G controls the clustering of the hues around the sectormean C of the template (greater σ clusters more, we use w/2)( p ) S ( p )H ' ( p) G( p) w(1 Gσ ( H ( p ) C ( p ) )2This may break up coherent regions intodisjointly colored regions to avoid this, may embed a graph-cuts basedlabeling into the shifting procedurenon-harmonicharmonizedfrom Cohen ‘08Color Harmonization: Examplefrom Cohen ‘08Color Harmonization: ExampleCollage harmonization (from Wang ‘08):Collage harmonization (from Cohen ’06):non-harmonicharmonized (Ttype)

Color ConstancyA psychophysical phenomenon: accounts for the ability of humans to accurately perceive the "color" of Color Constancy: Exampleilluminant Ailluminant Billuminant Can object under different lighting conditionslighting, or illumination, may vary both over a viewed scene and overtime yet the perceived color is constantin fact, constant illumination over a scene is almost neverencountered in real lifeGiven an object, the colors we perceive (within limits) remainthe same, even though the spectral content ("color") of sunlight varies greatly through the dayand with weather conditions artificial light sources also vary greatly fromone to anotherChromatic Aberrationfrom: J. Döllner, U PotsdamWhy Color? Color Adds More Dimensionsfrom: M. Stone

Color Adds Aestheticsfrom: M. StoneBut Mapping to Color Can Cause Problemsfrom: M. StoneColor Mapsfrom: Rogowitz/TreinishColor Map: Segmentation Tasksfrom: Rogowitz/Treinish

Color Map: Rainbowfrom: Rogowitz/TreinishColor Maps: Spatial Frequency IssuesColor Map: Linear Huefrom: Rogowitz/TreinishColor Maps: Low vs. High Frequencyweather modellow frequencyradar scanhigh frequencyfrom: Rogowitz/Treinishfrom: Rogowitz/Treinish

Color Maps: HighlightingBrewer ScaleNominal scales distinct hues, but similar emphasisSequential scales vary in lightness and saturation vary slightly in hueDiverging scale complementary sequential scales neutral at “zero”from: Rogowitz/Treinishfrom: M. Stone (see also colorbrewer.org)Brewer Scalesfrom: M. Stone (see also colorbrewer.org)Example for Proper Use of Color

ReferencesMaureen Stone, A Field Guide to Digital Color, AK Peters2003 color perception and design with colorColin Ware, Perception and Information Visualization: 2ndEdition, Morgan Kaufman, 2004 book specifically geared towards information visualizationBernice Rogowitz and Lloyd Treinish, “An architecture forperceptual rule-based visualization,” Proc. IEEEVisualization 1993, pp. 236-243, 1993 see dings/pravda/truevis.htmColor brewer: http://www.colorbrewer.org

Methods Global methods: scale each pixel according to a fixed curve the key issue is here: the shape of the curve Local methods: group small neighborhoods by their average value scale these averages down add detail back in Methods Local methods: group small neighborhoods by their average value scale these averages down add detail back in