February 2011

LianPu is an automatic face tagging system written in MATLAB. Similar to Picasa and iPhoto, given a natural photograph, faces appeared in that photograph will be tagged with names. The name is from Chinese "脸谱" (pinyin: Liǎn Pǔ), which means facial makeup patterns and drawings used to represent specific characteristics in Beijing Opera.

Social networking websites usually contain a huge amount of photographs uploaded by their users, but none of the sites provide automatic face tagging functionality. As a result, users have to tag the faces in the photos themselves. Face detection and face recognition are developed technology, but they have been actively used only for special purposes such as CCTV system, or special photos such as passport photos. We aim to develop a prototype system of integrated face detection and face recognition on natural photos that ordinary users would take and share. To test on real wild images, we build our own face database off these natural photos. Our system consists of two phases: face detection and face recognition. For face detection, we use Viola-Jones method, and for face recognition, grid-based local matching of SIFT features technique is employed. With limited training images in the face database we build, we achieve an online recognition rate of 48.7%, and an offline recognition rate of 82.2%.


Presentation Report


He Zhao, Lu Guoyu