stronger red component than other parts of the mouth [17]. Therefore,
we apply the following simple transformation on the mouth region to
convert the color image into gray scale [35]:
I ¼ R−2:4 G þ B: ð1Þ
The effect of this transformation is clearly shown in Fig. 2(c). The lip
region becomesmuch brighter than the background pixels of themouth
region.We then employ the Otsu binarization method [36] to segment
the lips in the enhanced image. The lip contour is obtained by dilating
the segmented lips with the following 3 × 3 morphological operator
se = [1 0 1; 1 1 1; 1 0 1]. The undilated segmented image is then
subtracted from the dilated one leaving only the lip contour, Fig. 2(d).
Figs. 2 and 3 show the whole of the lip contour extraction process
using exemplar face images from each of the three datasets. From
these examples, it is clear that our simple lip extraction approach
works very well, even on faces with extensive facial hair, as shown in