In this paper we have presented a classification scheme using the geometric moments as features. The hybrid system generates the features optically and performs the classification digitally. The system provides invariance to distortion and allows corrections for nonuniformities and nonlinearities in the optical system. For the moment feature set, we can describe analytically the probability density function set, we can describe analytically the probability density function for the moments and thereby derive an optimum classification scheme, eliminating the need for a training set. We have tested our classification system through digital simulation on real world ship images using reference feature vectors obtained only from models. The performance obtained was excellent and the preprocessing required only simple histogram and thresholding operations for image segmentation. These operations included producing row-projection histograms (from which the image regions were located) and statistical estimates of each region.
Distortion invariant recognition using a moment feature space
Verbesserungsinvariante Zielerkennung unter Verwendung von einem momentanen Ortsmerkmal
1983
4 Seiten, 7 Bilder, 5 Quellen
Aufsatz (Konferenz)
Englisch
Distortion Invariant Optical Pattern Recognition
British Library Conference Proceedings | 1993
|A Distortion-Invariant Pattern Recognition Algorithm
SPIE | 1987
|Distortion-Invariant Recognition via Jittered Queries
NTRS | 2000
|FUSION MOMENT INVARIANT RECOGNITION OF AIRCRAFT
TIBKAT | 2018
|Distortion-invariant pattern recognition with local correlations
British Library Online Contents | 2011
|