Generic Flaw Detection within Images

This paper looks at the problem of detecting aws or anomalies in greyscale images. A generic approach is adopted which uses little prior knowledge of the type of image. We assume only that most of the image will consist of`background', and that any anomaly will be relatively small in area. The methods explored involve extracting localised, but scale invariant features from an image and expressing them as a set of higher level entities, a process called the UpWrite. Once this has been achieved data points may be further UpWritten or simply classiied. This paper describes an eeective and generic method used to locate anomalies in images. This is demonstrated through examples using images varying in size, scale, intensity and features, but with no programming or parameter modiications. It is conjectured that the local processing performed here is a model for the behaviour of neurons in the visual system.