A technique is presented in which an improved and extended Hough transform (quasiparallel implementation) achieve independence of the number of original boundary points. Propagation of maximum evidence reduces the dimensionality of the accumulator space. A back-transformation of the complete accumulator space to the boundary points leads to surviving optimal parameter sets and uniquely assigns contour points to most evident parametric curves, which simplifies the interpretation substantially. The procedural steps are presented and illustrated for the case of circular object detection.