Optical EngineeringApplying the eigenfaces and Fisherfaces methodology to circuit-board inspection
|Format||Member Price||Non-Member Price|
This paper presents the application of two face recognition techniques to a manufacturing recognition problem, the detection of the presence of surface-mounted devices on printed circuit boards. Without assuming that the region of interest (ROI) is well framed, a preprocessing stage is developed to find and extract this subimage from a larger image using a series of image-processing steps. The eigenfaces and Fisherfaces feature extraction methods are then applied to the ROIs in order to project them onto lower-dimensional, optimal feature subspaces in which the classification is done. An important part of the work presented is the evaluation and selection of most discriminating eigenimage feature projections to use in classification. Experimental results on a testing database of 37,000 images yields 87% to 94% correctclassification rates for both feature extraction strategies.