Detection of a face by color is fast and reliable, but not always precise. Detection by blinking is precise, but requires capturing an image pair during a blink. Correlation can be used to complete these two techniques and to hold the face centered in the image as the head moves. The crucial question is how large a window to correlate (N) and over how large a search region to search.
We have found that a 7 by 7 search region extracted from the center of a face is sufficient to determine the precise location of the face in the next image. The size of the search region is determiend by the speed with which a face can move between two frames. Calling on the finger tracking results above, we optimised the search region to maximize the frequency of image acquisition. With our current hardware this is provided by a search region to 26 by 26 pixels, but must be adjusted when several image processes are running in parallel. Correlation can also be used in isolation to track a face. As an test of the robustness of correlation, we acquired a sequence of images of a head turning (shown in figure 6 below). We then correlated the center image from the sequence to produce the correlation graph shown in figure 7. The graph shows the results of zero-mean normalized correlation for the 40th face image with the other images in this set.
Figure 7: Correlation values for correlation of face number 40 with the
face images from figure 6.