| 102C5 |
|---|
| Face Detection: An Integrated Approach |
| AUTHORS |
| Masanori Sugisaka, Xinjian Fan (Oita University) |
| ABSTRACT |
| We present an approach to real-time face detection using multi-modal integration. Specifically, we combine a neural network based face filter, skin color, and genetic search into a single robust system. The neural network was trained with many face examples and non-face examples to get the concept of face or non-face. It receives a 20¡ß20 pixel region of the image as input, and generates an output ranging from 0 to 1, signing how close it is to a face. The task of finding a well-suited sub-image can be formulated as a discrete global optimization problem. An exhaustive search would produce the optimal result, but is not computationally feasible. We proposed to use GAs to search the image efficiently: each individual represents a sub-image within the input image; during evolution, the best individual which detection value is above a predefined threshold would be reported as a face. Skin color is another important feature of face images. We first build a probability distribution image of the skin color from the input image using a histogram based model. By punishing the sub-images with few skin color information with a low fitness, the genetic search can be conducted to the area only with skin color information and thus will greatly speed the search process. Experiment results showed the effectiveness of our integrated method. |
Back to Program
Time Table
Top of This Site
Annual Conference 2002