Accurate and Reliable Framework for Fast Parametric Curves Detection
Publication type
Conference paper
Author(s)
A. Levashov, D. Yurin
Publication date
November 2011
Conference
8th Open German-Russian Workshop “Pattern Recognition and Image Understanding” (OGRW-8-2011)
Page(s)
178-181
Address
Lobachevsky State University of Nizhny Novgorod, November 21-26
Abstract
We propose a multi-stage framework for parametric curves detection in images. Firstly, we search edge curves, then, after several preparatory procedures, they are stored as chains of connected pixels. These chains are analyzed both as whole and as piecemeal. Analysis of each fragment firstly uses randomized methods to reject certainly inappropriate models. Only for remaining hypothesis we perform accurate model parameter estimation by least squares method and check reliability of a hypothesis by chi-square criterion. Such two stage approach allows us achieve high performance, search parametric curves of different types simultaneously, including cases when different fragments of one curve correspond to different models. The measurement results of parameters estimation accuracy and algorithm operation time are given.
