Face detection with abstract template

In this paper an abstract template is constructed for extracting facial information from images. Different from the traditional template, which only includes color or gradient information in pixel level, abstract templates synthesizes texture, shape and scale information of the objects with relative parameters. This parameterized model can be used to describe the object to be recognized together with a measure of how like it fits the image data. Three kinds of abstract templates are proposed in eye, eye pair, and face texture detection separately for face detection. After applying the abstract template, the parameters computed can be used for object description and recognition.