Complex Autoregressive Model for Shape Recognition

The problem of fitting a straight line to a planar set of pointsis &considered.; A parameter space computational approach capable offitting one or more lines to a set of points is presented. The suggestedalgorithm handles errors in both coordinates of the data points, evenwhen the error variances vary between coordinates and among pointsand can be readily made robust to outliers. The algorithm is quite genera1and allows line fitting according to several useful optimality criteria tobe performed within a single computational framework. It is observedthat certain extensions of the Hough transform can be tuned to beequivalent to well-known M estimators, thus allowing computationallyefficient approximate M estimation.