A new robust real-time method for extracting human silhouettes from color images

A new method for extracting human silhouettes from color images in real-time regardless of changes in lighting conditions and backgrounds is proposed. In the method, RGB values of each pixel in color images acquired by a CCD camera are converted to YIQ values. Before extracting human silhouettes, the mean YIQ values of each pixel and the threshold values for the Y, I, Q components are obtained in advance from the image sequence acquired by the camera that observed the background. Based on the mean values and the threshold values, each pixel is judged whether it belongs to the human silhouette or not. The proposed method was used with the virtual metamorphosis system, in which anyone can change his or her form to any other form, and the robustness and real-time-ness were confirmed.