Bridging Shape and Reflectance Presented by Jongwoo Lim Feb 18 2003 Bridging Shape and Reflectance p 1 23 Rendering vs Inverse Rendering Geometry Rendering Radiance Maps Reflectance Property Lighting Geometry Inverse Rendering Reflectance Property Radiance Maps Lighting Bridging Shape and Reflectance p 2 23 Parametric BRDF s L B I Radiance BRDF Irradiance Torrence Sparrow model s 2 2 2 BT S i o d s d cos i e cos o Ward model isotropic version d s tan2 2 BW i o d s e 2 4 cos i cos o Bridging Shape and Reflectance p 3 23 Renderings with Ward BRDF Isotropic Anisotropic from Simon Premoze s homepage http www cs utah edu premoze brdf Bridging Shape and Reflectance p 4 23 BRDF Estimation We need to estimate BRDF at every point on the object Knowns Unknowns Radiance L Irradiance I Surface normal n BRDF parameters d s Light source direction s Viewing direction v Bridging Shape and Reflectance p 5 23 Difficulties in Estimation Estimation of specular parameters is difficult The specular lobe is highly peaked it is difficult to observe specular points in images The specular term is not linear it requires many samples to estimate the parameter Bridging Shape and Reflectance p 6 23 Reflection Component Separation We can infer the contribution of diffuse and specular part MR MG MB K o1 1 cos i1 D R S R cos iN K oN N t KD GK G D GS t KS D G S G D B S B 150 0 intensity red green blue 100 0 50 0 0 0 55 65 150 0 intensity M 75 85 95 105 image frame diffuse red diffuse green diffuse blue specular red specular green specular blue 100 0 50 0 0 0 55 65 75 85 95 105 image frame Bridging Shape and Reflectance p 7 23 Illumination Condition Direct Illumination all reflected lights are from light sources only one reflection light source surface camera Global Illumination reflecting surfaces also work as light sources multiple reflections until being observed Bridging Shape and Reflectance p 8 23 Inverse Radiosity Lambertian Assume Lambertian surfaces L i E i i X Aj L j F ij Lj F ij j Li radiance radiosity Ei emission light source i albedo BRDF Fij pi Ei C form factor between patches i L i E i Li X L j F ij j Bridging Shape and Reflectance p 9 23 Inverse Radiosity Parametric BRDF radiance emission diffuse specular L Cv Pi E Cv Pi X LPi Aj F Pi Aj d j s X L Pi A j K C v Pi A j Aj LCk A j L PiA j pi L Cv Pi Ck Cv j Bridging Shape and Reflectance p 10 23 Inverse Radiosity Parametric BRDF Radiance from one point can vary with viewing directions LPi Aj L Ck Aj SCk Aj SPi Aj L Ck Aj SCk Pi Aj Iteratively estimate S with initial guess S 0 Bridging Shape and Reflectance p 11 23 Inverse Global Illumination Algorithm Assume constant BRDF over each patch Detect specular highlights on surfaces geometrically Choose sample points inside around each highlight Build links between sample points and patches Assign L0 with average radiance value and S 0 Iterate Update L using S of each link Optimize each surface s BRDF parameters Estimate S with new BRDF parameters Bridging Shape and Reflectance p 12 23 Monte Carlo Sampling Aj Q PiAj L PiAj QC k Aj Pi Ck Cv One bounce approximation is enough for this purpose Bridging Shape and Reflectance p 13 23 Diffuse Albedo Map Assume constant specular property over each patch Dif f use x d x Irradiance x Dif f use x L x Specular x Give lower weight on sample which has large specularity whose viewing angle is grazing the surface Bridging Shape and Reflectance p 14 23 Geometry So far we assumed that the geometry is already given input by hand measure from the object laser range finder stereo vision shape from X light stripe range nder object color camera light source robotic arm Bridging Shape and Reflectance p 15 23 Geometry Acquisition Procedure 1 Surface acquisition from each range image 2 Alignment of range images 3 Retrieving 3D representation Merging based on a volumetric representation Isosurface extraction a range image acquisition b alignment c merging d isosurface extraction Bridging Shape and Reflectance p 16 23 Surface Normal Estimation Why is the surface normal important BRDF is evaluated at each surface points locally For accurate estimation precise are required The eigenvector of the covariance matrix of neighbor points principal axis X x x x x t n N ull neighbor Bridging Shape and Reflectance p 17 23 Experiment Setup Conference Room Bridging Shape and Reflectance p 18 23 Result Conference Room a Initial hierarchical polygon mesh with radiances assigned from images b Synthetic rendering of recovered properties under original illumination Bridging Shape and Reflectance p 19 23 Result Conference Room c Synthetic rendering of room under novel illumination d Synthetic rendering of room with seven virtual objects added Bridging Shape and Reflectance p 20 23 Result Whiteboard Bridging Shape and Reflectance p 21 23 Result Mug Synthesized object images input frame 50 synthesized input frame 80 synthesized Bridging Shape and Reflectance p 22 23 Reference Y Sato M D Wheeler K Ikeuchi Object Shape and Reflectane Modeling from Observation SIGGRAPH 1997 Y Yu P Debevec J Malik T Haukins Inverse Global Illumination Recovering Reflectance Models for Real Scenes from Photographs SIGGRAPH 1999 P Debevec J Malik Recovering High Dynamic Range Radiance Maps from Photographs SIGGRAPH 1997 P Debevec Rendering Synthetic Objects into Real Scenes Bridging Traditional Image based Graphics with Global Illumination and High Dynamic Range Photography SIGGRAPH 1998 Bridging Shape and Reflectance p 23 23
View Full Document
Unlocking...