Trainable method of parametric shape description

Shape models have been used widely to achieve robust interpretation of complex images. They allow image evidence to be organized into plausible interpretations which can then be verified”“. We are interested in the class of problems where shapes are variable. Important examples are the inspection of complex manufactured assemblies where relative motion between subparts is possible. and medical image interpretation where biological variation is present. In such applications it is Senerally the case that some aspects of shape arc Invariant whilst others are subject to constrained variability. The problem of adequately modelling such behaviour in ;I general way has not been solved. We have previously described a method of shape representation based on modetling the statistical distributions of chord lengths between control points placed in a consistent manner on each shape in a training set “.I ‘. .I‘he objective\ of the work we describe here were to significantly develop this basic idea to automatically: