Nick Cerneaz and Mike Brady
Identification of the blood vessels and milk ducts within an X-ray mammogram image allows intelligent analysis of calcification clusters and mmass borders detected by other means. These vessel ligatures are however generally buried within noise with variance of the order of the vessel signal it- self. We present a scale-matched feature detector designed to extract nominally 1-dimensional ridges from an image with a signal-to-noise ratio approximatlng unity. The algorithm uses first and second difference gradient measures of the original nonsmoothed images to drive a search exploiting the priori knowledge of the expected vessel (ridge) features. The algorithm is illustrated with an ex~mle.