A Hough transform technique for detection of rotationally invariant surface features

The Hough transform is a relatively simple and robust method for detection of economically parameterized geometric features via accumulators in the discretized parameter space. Several methods for detecting circular structure have been proposed, with centers of circles often detected as intersections of normals to the tangents of curves. We modify the transform to detect general rotationally invariant, continuously-valued image features, based on gradient measurements independent of radii. This method is robust to uncertainties in surfaces which are poorly modeled by standard stochastic models. Examples of application of the transform are from paleontological classification of digitized fossil molar surfaces.<<ETX>>