Highly Discriminative Invariant FEatures for Image Matching

Abstract. In this paper, we present novel image-derived, invariant features that accurately capture both the geometric and color properties of an imaged object. These features can distinguish between objects that have the same general appearance (e.g., dierent kinds of sh), in addition to the typical task of distinguishing objects from dierent classes (e.g. sh vs. airplanes). Furthermore, these image features are insensitive to changes in an object's appearance due to rigid-body motion, ane shape deformation, changes of parameterization, perspective distortion, view point change and changes in scene illumination. The new features are readily applicable to searching large image databases for specic images. We present experimental results to demonstrate the validity of the approach, which is robust and tolerant to noise.

• Type: Conference
• Year: 1999
• Where: VISUAL
• Authors: Ronald-Bryan O. Alferez, Yuan-Fang Wang
• Keywords: Ane Shape Deformation, Di Erent Classes, Di Erent Kinds, Information Management, VISUAL 1999