A probabilistic decision-based neural network for locating deformable objects and its applications to surveillance system and video browsing

Detection of a (deformable) pattern or object is an important machine learning and computer vision problem. The task involves nding a speciic (but locally deformable) patterns in images, such as human faces and eyes/mouths. They have many important commercial applications including ATM, access control, surveillance, video conferencing, and video libraries. Therefore, it has attracted much attention in recent years. This paper presents a decision-based neural network for nding such patterns with speciic applications to detecting human faces and locating eyes in the faces. The system built upon the proposed has been demonstrated to be applicable under reasonable variations of orientation and/or lighting, and with possibility of eye glasses. This method has been shown to be very robust against large variation of face features and eye shapes. The algorithm takes only 200 ms on a SUN Sparc20 workstation to nd human faces in an image with 320x240 pixels. For a facial image with 320x240 pixels, the algorithm takes 500 ms to locate two eyes on a SUN Sparc20 workstation. Furthermore, the algorithm can be easily implemented via specialized hardware for real time performance. We have applied this technique to two applications (surveillance system , video browsing) and this paper provides their experimental results. Although we have only shown its successful implementation on face detection and eye localization, the proposed technique is meant for more general applications of detection of any (locally deformable) object. Since there are many other object detection problems in real world applications that share similar problem characteristics; therefore , the current application systems have provided valuable insight on how one might approach other deformable pattern or objects.