Face Recognition by Support Vector Machines
FG '00 Proceedings of the Fourth IEEE International Conference on Automatic Face and Gesture Recognition 2000
Hybrid Hidden Markov Model for Face Recognition
SSIAI '00 Proceedings of the 4th IEEE Southwest Symposium on Image Analysis and Interpretation
Facial Component Extraction and Face Recognition with Support Vector Machines
FGR '02 Proceedings of the Fifth IEEE International Conference on Automatic Face and Gesture Recognition
Geometric and Optical Flow Based Method for Facial Expression Recognition in Color Image Sequences
ICCVG 2008 Proceedings of the International Conference on Computer Vision and Graphics: Revised Papers
Proceedings of the International Conference & Workshop on Emerging Trends in Technology
Hi-index | 0.00 |
In this paper, we present a new, real time solution for face detection. The technique applies a combination of skin region color ratios and K-Mean clustering. All code was written in fully interoperable Java. The process of detecting a face from a video sequence in real time is essential in applications such as human surveillance, human computer-interaction, and for further face recognition research purposes. The face detection process is divided into four phases: Video Database Acquisition (VDA), Frame Sequence Extraction (FSE), Skin Region Detection (SRD), and K-Mean Face Segmentation (KFS). The MPEG formatted videos are converted to JPEG images depending on the user specified frame rate (FSE phase). During this conversion, the face detection process itself (SRD and KFS phases) runs on each of the images as they are being converted. Skin regions are detected in the images, which act as the input for the K-Mean Face Segmentation phase. The algorithm was tested on 18 video sequences (acquired using a cost effective commodity DV camera), and was effective regardless of age, gender, size, race, or skin tones. The detection algorithm also handled varying illumination conditions, such as bright sunlight, sufficient light, and dim light conditions (with R, G, and B values only in the teens and high single digits), and is pose invariant. The time taken to detect and store the normalized faces was comparable to the length of the video, and in some cases it was even less. This system works in True Real Time (TRT).