A real-time multiple human face detection and recognition based on scalable vocabulary tree

Abstract

Human face detection and recognition have many applications such as camera surveillance,

biometric applications, and human computer interface. Automatic human

face detection and recognition from images is a challenging task due to the uncertainty

in cluttered background, view point, illumination, occlusion, and facial

aging.

An e cient implementation of real time face detection has been done. Face detection

is based on sophisticated lters. The detection accuracy of the state-of-the-art object

detection algorithms is rather high. However, the computational cost is also high.

In this work, the face lter is trained o ine from the available source code of general

object detecion algorithm. The main idea of the algorithm is object detection based

on mixtures of multiscale deformable part models. Once the face lter is trained,

it is used in the C code for real-time video surveillance system. Using this system,

webcam can capture face of humans in realtime. The face detection accuracy is

tested to be over 90% in 2 seconds. In contrast, using the original MATLAB code,

the detection accuracy is the same while the detection time ranges from 15s to 30s.

An e ective algorithm for real time face recognition has been applied. According to

comprehensive survey of the recognition methods, it is found that Scalable Vocabulary

Tree (SVT) is e ective in general image recognition. As compared to general

scene images, which contain generic topics such as outdoor, o ce, and sky, face

images usually contain speci c face parts such as eye, nose and mouth. The implementation

is carried out in MATLAB since the algorithm is rather e cient. The

face detection accuracy is also roughly 90% in less than 0.5s.In addition, new face images have been collected. In order to test the performance

of face detection, the author has been using the webcam to capture her face based

on the developed software since half year ago. The capturing conditions varies,

including di erent lightings, angles, occlusions, makeups, hairstyles, glasses, etc.

Then the detected faces are recognized using the face image databases.