The Use of Symmetry Chords for Expressing Grey Level Constraints

In many object recognition systems only geometric constraints on the boundaries of the objects are used. Many authors have shown the merit in determining the distinctiveness of boundary constraints, and using these to speed up and improve the robustness of object recognition algorithms. However, if the objects have variable shapes or the image is cluttered, then the use of boundary constraints alone can cause object recognition to be be both slow and lacking in robustness. We present a method for expressing grey level constraints using the grey levels along symmetry chords. We describe a method for forming groups of symmetry chords, and calculating their distinctiveness on the basis of the grey levels along the chords in training images. We show how distinct groups of symmetry chords can be used in object recognition by creating evidence images for specific groups of boundary points. Our initial results are promising.