torecognitionof3-Dobjects-PatternAnalysisandMachine.PDFVIP

640 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 12, NO. 7, JULY 1990 Application of Affine-Invariant Fourier Descriptors to Recognition of 3-D Objects KLAUS ARBTER, WESLEY E. SNYDER, SENIOR MEMBER, IEEE, HANS BURKHARDT, AND GERD HIRZINGER Abstract-In this work, the method of Fourier descriptors has been normalization was initially developed by Granlund [7] and extended to produce a set of normalized coefficients which are invari- later extended to address the issue of completeness [3], ant under any affine transformation (translation, rotation, scaling, and shearing). The method is based on a parameterized boundary descrip- PI. tion which is transformed to the Fourier domain and normalized there The class of affine transforms includes the similarity to eliminate dependencies on the affine transformation and on the start- transforms, but in addition allows “shear. ” Under shear, ing point. the shape of the object boundary will no longer be pre- Invariance to affine transforms allows considerable robustness when served, as illustrated in Fig. 1. Such shape distortion can applied to images of objects which rotate in all three dimensions. This is demonstrated by processing silhouettes of aircraft as the aircraft typically arise if a planar object is observed by a camera maneuver in three-space. under arbitrary orientations with respect to the plane. In this case, the perspective transformation would be the Zndex Terms-Affine transformation, features, Fourier descriptors, most appropriate, however, the affine transformation may invariants, shape, 3-D parameter estimation, 2-D parameter determi- nation. be used as an approximatio