Boundary Pattern Analysis
Recognition of binary objects by boundary pattern analysis should be a straightforward process, but as this chapter shows, a number of problems need to be overcome. In particular, any boundary distortions such as those due to breakage or several objects being in contact may result in total failure of the matching process. This chapter discusses the problems and analyzes possible solutions.
• an analysis of hysteresis thresholding procedures.
• the centroidal profile approach and its limitations.
• how the method may be speeded up.
• how recognition based on the (s, ψ) boundary plot is significantly more robust.
• how the (s, ψ) plot leads to the more convenient (s, κ) plot.
• the exact relation between k and ψ.
• more rigorous ways of dealing with the occlusion problem.
Disparate ways are available for representing object boundaries and for measuring and recognizing objects using them. All the methods are subject to the same ultimate difficulties—particularly that of managing occlusion (which necessarily removes relevant data) and that of inaccuracy in the pixel-based description. Sound ways of managing occlusion are suggested in Section 7.7. This work presages Part 2 of this volume where the Hough transform is employed widely for object recognition—albeit not by overt boundary tracking, but by an altogether more robust parallel processing approach.