A Hidden Markov Model object recognition technique for incomplete and distorted corner sequences

摘要

This paper presents a new technique for planar object recognition based on Hidden Markov Models. First, the contour of the object is processed to extract a sequence of high curvature points. These points are extracted from a new adaptively extracted curvature function which is resistant against noise and transformations. Each corner is characterized by its subtended angle and its distance to the next corner. Then, corner sequences are analyzed by using HMMs. The method has been successfully tested for different databases. Its main advantage is that it can deal with incomplete and distorted corner sequences.