Object detection and localization by dynamic template warping

A simple method is presented for detecting, localizing and recognizing inst ances of classes of objects, while accommodating a wide variation in an obj ect's pose. The method utilizes a small two-dimensional template that is wa rped into an image, and converts localization to a one-dimensional sub-prob lem, with the search for a match between image and template executed by dyn amic programming. For roughly cylindrical objects (like heads), the method recovers three of the six degrees of freedom of motion (2 translation, 1 ro tation), and accommodates two more degrees of freedom in the search process (1 rotation, 1 translation). Experiments demonstrate that the method provi des an efficient search strategy that outperforms normalized correlation. T his is demonstrated in the example domain of face detection and localizatio n, and can extended to more general detection tasks. An additional techniqu e recovers rough object pose from the match results, and is used in a two s tage recognition experiment in conjunction with maximization of mutual info rmation.