Abstract
There are many products such as computers or mobile telephones whose operation buttons or keys are printed with letters. The quality of these printed letters are visually inspected in most cases before shipment to find the defects such as hollows or soils. However, it is difficult to clearly define and quantify the defects because of the defect type, location, size and human error, causing disputes between the buyer and the seller. The problem has not been solved yet. It is therefore necessary to develop an automated inspection method which can quantify the quality of printed letters. In this study, Mahalanobis-Taguchi System was used to construct the Mahalanobis space from normal data. Normally printed letters were pre-selected by inspectors and the Mahalanobis space was calculated from the characteristics which were extracted from letter images. The experimental results agreed well with the results of human inspection.
