S.A. Suandi, K. Gohara, and T. Ejima (Japan)
human-computer interaction, TSL, tracking facial features
A fast image processing technique to extract human face and track human facial features in a video sequence is in troduced in this paper. This system can be used, besides as an online face tracker system, as a multimedia appli cation, online teleconferencing, a basic system to study human behaviour etc. In this paper, we introduce a sys tem that extracts human face based on TSL colour space and tracks the eyes and mouth based on the information yielded from discriminate analysis method. At inertial state, TSL colour space is used to find skin colour pixel from the input rgb image. Considering only the chromi nance elements of each pixel, a system that robust to lu minance change is shown in this paper. With only pixels that have skin colour attribution in TSL colour space, we form face region candidates by connecting pixels that are 4-neighbour connected. When a face region candidate exists, using horizontal Haar Wavelet transformation, we extract horizontal parts on face. We then compute the vertical and horizontal integral projection of this image. For each projection, we find the threshold using discrim inate analysis method. Using these information, our sys tem finds and tracks the facial features, namely mouth and eyes. We also show that the system can also estimate face pose by using the same information.
Important Links:
Go Back
