US Patent 7440607 Outlier substrate inspection

A method of detecting anomalies in a test image. Test features of pixels within the test image are selected, and reference features of pixels within at least one reference image are also selected. A signal distribution of test features and reference features in a multi-dimensional feature space is created, and stored. Those test features of the test image that do not satisfy a set of criteria for normalcy are selected as candidate points. Those candidate points that are statistical outliers are identified as anomalies. Positions of the anomalies are located using the stored signal distribution within which the defects have been identified as a lookup table.