Ever since the character strings on silicon wafers have been read using OCR cameras, there has been a problem with damaged characters. This problem is due to reflection from the light source or the physical damage of the characters themselves. There are obvious types of damage that occur frequently on many of the bitmaps that the OCR camera reads. With these types, one can test them to find the most damaging types on each particular character that has occurred. However, currently there is no known research that systematically determines the worst damages or limits of damage to characters for specific OCR methods such as template matching or neural network algorithms. This paper presents algorithms for testing common forms of damages on template-matching optical readers reading strings on silicon wafers. It also displays results from combining a simple neural network and the algorithms. The results on readability study are critical for the development of robust OCR systems.
展开▼
