Christopher Coelho, Aaron Heller, Joseph L. Mundy, David A. Forsyth and Andrew Zisserman,
An Experimental Evaluation of Projective Invariants,
In Joseph Mundy and Andrew Zisserman eds., Geometric Invariance in Computer Vision,
Cambridge, MA, USA, Series: Artificial intelligence, pp. 87-104, The MIT Press, 1992.Projective invariants are being used in vision applications with increasing frequency. This paper addresses the question of reliability of five commonly used planar projective invariants given the types of errors encountered in actual applications. They are analyzed from a theoretical viewpoint to derive expressions that characterize the propagation of error and from an experimental viewpoint by constructing test patterns with known values of the invariants and comparing these to the values calculated from image data. We discover that the accuracy and stability of the invariants are directly dependent on the performance of the feature extraction scheme employed to recover the test pattern geometry from the images. Because of this, we are able to devise a feature extractor testing methodology based on planar projective invariants. Preliminary results from tests on three different feature extractors are presented.