Identification of human face profiles by computer

摘要

Human face profiles are automatically recognized by computer processes. Artists' tracings of profile photographs are scanned into a minicomputer. Fiducial marks, automatically placed, serve as key loci for establishing geometric measures such as distances, angles, and areas. “Features” established by these measures comprise 10-dimensional vectors which uniquely represent the face profiles. “Unknown” feature vectors are compared to stored files of vectors to determine similarity and to establish identification. Euclidean distance measures are supplemented by file-reducing techniques based on a priori similarity measures which eliminate unlikely candidates. Absolute recognition accuracy ranges from 86 to nearly 100% under various conditions for a population of 121 persons having 3 poses each on file.