Most image processing methods require the image to be in digitised form and where the image is processed by comparing pixel values in a predetermined mask. If there is an error or distortion (which at some level there will always be) then this can adversely affect the processing of the image.
This invention uses software, developed using a new mathematical theorem (convexity-based fast and effective global geometric transforms), to smooth images and to identify points on an image or a data array which are geometrically singular. The various methods for identification of such singular points enable, for example, restoration of an image by replacement of the missing or distorted parts, selective enhancement of features of an image or a data array, and detection of edges, corners, multi-scale medial axis, crossing points and end points. The method is fast and, unlike many alternative methods, is highly stable. Typical applications include image reconstruction, digital watermarks, medical imaging, and image analysis.
Patent application is pending.
Inventors: Elaine Crooks, Antonio Orlando & Zewei Zhang