Abstract: The present invention is a hybrid technique for finding defects on digitized device images using a combination of spatial domain and frequency domain techniques. The two dimensional spectra of two images are found using Fourier like transforms. Any strong harmonics in the spectra are removed, using the same spectral filter on both spectra. The images are then aligned, transformed back to the spatial domain, and subtracted. The resulting spectrally-filtered difference image is thresholded and analyzed for defects. Use of the hybrid technique of the present invention to process digitized images results in the highest-performance and most flexible defect detection system. It is the best performer on both array and random devices, and it can cope with problems such as shading variations and the dark-bright problem that no other technique can address.

Abstract: A method of inspecting repeating pattern devices according to which an image of the patterns is aligned with an array of pixels in the image detection plane of an optical detector. The image is magnified to a scale so that features of patterns repeated in the image occupy corresponding pixels or groups of pixels repeated in the array. Data is resolved from selected pixels and directly compared either to data obtained from corresponding pixels or from a data base, whereby defective features are identified through well-known data comparison techniques.