Compensation of non-uniform illumination and background variations using recursive least squares algorithm

Abstract

Illumination plays an important role in vision inspection tasks. Since incident light intensity at a point depends on its distance from the light source, perfectly uniform illumination can only be obtained with either a point source at infinite distance or an infinitely large planar source. For low-contrast images, most practical illumination methods such as incandescent lights and flourescent lights cannot provide illumination uniform enough for subsequent defect detection. In this paper, the recursive least squares estimation method is used to model the two-dimensional intensity surface of the incident light and defect background. The estimated intensity surface can then be used to normalize the raw images by sample subtraction without reducing the defect contrast. Dopant wafers with cosmetic defects such as gray spots, gray streaks and black spots present a particularly difficult task for general purpose vision systems. Because of the low contrast of these defects, common noise filtering methods such as median operator and edge detection algorithms such as Sobel operator often either fail to detect these defects or generate unacceptable false detections. The proposed background compensation method has been applied to these low-contrast dopant wafers and its effectiveness has been demonstrated.