Multi-face Detection System in Video Sequence

Abstract

This paper presents an improved system for detecting multiple faces in a video sequence where detection is not limited to frontal view. We propose an adaptive selection of skin models from two different ones in RGB and ratio rgb space to overcome the illumination problem caused by automatic focus of camera. With the proposed solution, we receive more reasonable skin detection for different human races. We modify local binary pattern (LBP) by adding a set of spatial templates. This LBP considers both principal local shapes and spatial textures of facial components. Human face is represented by a histogram of LBP coefficients. Moreover, the grayscale image of human face is changed to discrete cosine transform (DCT) coefficients used in embedded hidden Markov models (eHMMs). A modified LBP (mLBP) histogram matching and eHMMs are composed to hierarchical classifier to determine whether skin regions are faces or not. The results of our system on different cameras are presented and discussed. We compare the performance capability of our system with other systems using separately eHMMs or LBP histogram matching. The proposed system can work well to detect attitudes of faces like frontal, rotated and profile faces.