Discriminative Distance Measures for Image Matching

: Significant progress has been made by the computer vision community in recent years along two fronts: (i) developing complex spatial-temporal models for object registration and tracking, and (ii) applying pattern classification techniques to object detection and recognition. These two approaches are starting to converge for cases when the object appearances are known apriori in the form of a training set of images, and the standard assumption is that the training distribution is reasonably representative of the test distribution. However, in cases when the object appearances are unknown or suffer heavy bias in the test set, pattern classification techniques are typically discarded in favor of standard but somewhat ad-hoc matching measures such as SSD, histogram difference measures, Hausdorff and chamfer distances. This situation applies to problems as diverse as image mosaicing and tracking. In this talk, I will propose a principled approach to constructing the best matching distance...

• Type: Conference
• Year: 2004
• Where: ICPR
• Authors: Tat-Jen Cham, Xi Chen
• Keywords: Ad-hoc Matching Measures, Computer Vision, Histogram Difference Measures, ICPR 2004, Pattern Classification Techniques