Tibor Takács, László Vajta
Novel image similarity measurement in Automated Optical Inspection
Image similarity measurement is one of the most important topics in industrial image processing systems. In automated optical inspection (AOI) in electronic device manufacturing, the widely used methods are built mainly on separated analysis of gray-scale images, and do not apply the high similarity between captured pictures.
This paper presents a new methodology to measure the relative similarity of AOI images. Our method utilizes and satisfies the special conditions and requirements of AOI systems. The need for human intervention (parameter adjustment, calibration) is almost totally eliminated. As our experiments show, our novel techniques classify more than 98% of images in the perfect classes which makes the techniques built on this similarity measurement method entirely useful in industrial applications already at this stage of our research.