Elastic Window for Multiple Face Detection and Tracking from Video

Abstract

This paper deals with an efficient method for the detection and tracking of multiple moving faces from a video sequence. Appropriate detection of multiple faces from a video sequence is a challenging task due to the different combination of noise, illuminations, pose, and locations of the human face which is likely to differ from one frame to another. This paper presents a unique technique for multiple face detection from a video sequence. In this study, our major objective is to detect and track locations of multiple faces from video using elastic window. Additionally, the face tracking system includes the tracking of face motion. Firstly, for each pixel, local entropy is calculated by considering a $3 \times 3$ window for detecting the face edges. Subsequently, Gaussian filtering technique is used to eliminate the undesired edges. In this context, it may be noted that a video frame passes through a number of preprocessing steps in order to eliminate the background noise to realize the thin binary image consisting of face boundaries. The human face from video sequences can be tracked by calculating the scalar and vector distances of four corner points between two adjacent frames. The movement of corner points represents the position and location change of the face in the upcoming frame. The presented method has been tasted on several video database and obtained efficient detection and tracking of multiple faces from the video sequences.