| Summary: | 碩士 === 元智大學 === 工業工程與管理學系 === 90 === Pattern matching has been an important technique in machine vision for the applications of optical character recognition (OCR), object detection, motion analysis, and defect detection. Normalized correlation is the most common measure used for pattern matching. However, the traditional normalized correlation is computational intensive, and is sensitive to environmental changes. This prohibits normalized correlation for industrial inspection applications. In this research, we propose a method to improve the effectiveness and efficiency of the normalized correlation for defect detection application.
In order to reduce the variation of normalized correlation affected by the factors such as image displacement and intensity variation, a Gaussian smoothing filter is used to smooth both the reference image and scene image so that the evaluated correlation values can be stable respect to minor environmental changes. To improve the computational efficiency of the normalized correlation, a sum-table approach is applied, which makes the computation of normalized correlation invariant to the window size of two compared images. Experimental results on various industrial samples such as PCB, SMT and BGA have shown that the proposed method is very efficient and effective for defect detection application.
|
|---|
