Real-Time Face Recognition from Surveillance Video - Intelligent Video Event Analysis and Understanding - page 166

Fig. 5 Skin Detection

3.1.1 Skin Detection

Viola-Jones operates on greyscale images, but for our implementation, we decided
to also exploit the colour information in the video stream 5 . By applying skin colour
detection to the input video frames, it is possible to divide the search space into face
candidate regions and background regions. Face detection need only be applied to
face candidate regions. The segmentation of skin regions based on luminance and
chrominance values is discussed in [10].
O PEN CV's Adaptive Skin Detector was used to detect face candidates (see
figure 5). Skin-coloured areas (with a margin added around the edges to allow for
areas of shadow) were passed to the face detector. This resulted in a significant
speedup and also improved accuracy, by discarding face-like regions which are not
skin-coloured (false positives).
Another benefit of using the skin detector was to segment faces in profile view
from the background. This enabled more accurate detection of the facial profile,
overcoming the problem of detecting an outline against a textured background (see
above).
The results of skin detection are discussed in detail in section 5.1.

5

Indeed, the original Viola-Jones paper [31] suggests that alternative sources of information
can be integrated with the basic approach, to achieve even faster processing and better
frame rates.