Comparing geometric with normalized correlation pattern-matching techniques

Comparing geometric with normalized correlation pattern-matching techniques

Fernando Serra

Machine-Vision Group Manager Imaging Technology

Bedford, MA 01730

For equipment that requires precise, automated alignment, such as wafer sorters, ion-implantation systems, wafer steppers, and wire-bonding equipment, the most important challenge is locating reference patterns despite changes in a material`s appearance. To solve these pattern-location issues, geometric techniques maintain high precision despite changes in object/pattern size, orientation, shape, focus, contrast, or partial occlusion. These new techniques automatically adapt to the process variations encountered during the manufacturing process while still maintaining up to a 1/40-subpixel accuracy and repeatability.

But while geometric search techniques are being applied to complex machine-vision applications, normalized gray-scale correlation (NGC) still has its place. Indeed, for those OEMs using pattern-finding/search tools, the majority are using NGC techniques. This usage has evolved because commercially available vision tools that use geometric techniques are new to the market.

Geometric search techniques, such as Imaging Technology`s Smart Search, have been developed specifically to overcome difficult vision issues, issues that cannot handle or are poorly handled by gray-scale correlation, such as pattern occlusion, missing information, degraded images, rotation, nonlinear reflectance, and contrast reversal. However, in considering a typical machine-vision application, it is important to realize that geometric techniques are only a portion of the search process, that portion that locates the pattern under extreme conditions.