【正文】
CVPR04 Stockman (27 June 04) 1 Point/feature matching methods 1 Gee Stockman Computer Science and Engineering Michigan State University CVPR04 Stockman (27 June 04) 2 Why match points across images? ? For fusing information from different sensors or different times ? To detect changes in a scene from time t1 to time t2 ? To mosaic overlapping images of a scene to form one large image ? to recognize an object by registering its image to a map or model CVPR04 Stockman (27 June 04) 3 High level view: match what? ? image points I[x,y] to map or model M[u,v] ? image points I1[x,y] to image points I2[u,v] ? model points M1[x,y] to model points M2[u,v] CVPR04 Stockman (27 June 04) 4 Problems ? How to find landmarks to match across two images? ? How to distinguish one landmark from another? ? How to match N landmarks from I1 to M landmarks from I2, assuming that I2 has more and/or less than I1 ? CVPR04 Stockman (27 June 04) 5 More problems ? The mapping between the two image spaces may be nonlinear/plicated ? There may be errors in the locations of detected point or area features ? There may be extra features in one image or the other ? There may be missed features in one image or the other CVPR04 Stockman (27 June 04) 6 Some salient face points Points identified by surface curvature in neighborhood and location relative to other salient points. No R eye or R ear No L eye CVPR04 Stockman (27 June 04) 7 Salient points useful for matching sensed and model surfaces 3D model made by mosaicking multiple views images of face from Minolta Scanner CVPR04 Stockman (27 June 04) 8 Why match faces? ? Image to model for recognition or verification ? Image to model for cloning/animation (. animating a jackass in Shrek) ? Image to general model for detecting a face ? Image to image for recognition or verification CVPR04 Stockman (27 June 04) 9 Matching landmarks via cross correlation Treat image neighborhoods N1 and N2 as functions or vectors and match them using