Face detection is a preprocessing step for facial recognition systems. The goal is to localize faces in digital images and video in real-time. Skin is an important element for detecting images containing skin or skin-like regions, as skin covers most of the face and varies less in color than brightness between people. The researchers used a mathematical model of skin color represented in three color spaces (rg, HSB, and YCbCr) generated from face images. Skin was modeled using mean and covariance characteristics of chromaticity from training images. Face detection was performed by calculating the probability that each pixel belonged to the skin distribution model in the color spaces. Experiments were conducted on over 3,000 face images from seven databases. Accuracy needs improvement
Mathematical Model of Skin Color for Face Detection
1. Mathematical Model of Skin
Color for Face Detection
Setiawan Hadi, Adang Suwandi A,
Iping Supriana S, Farid Wazdi
Universitas Padjadjaran, Bandung, Indonesia
Institut Teknologi Bandung, Indonesia
4. Skin-based Face Detection
• Skin important element in detecting image
that contain skin or skin-like region
• Skin is special
– covers most of the face image area
– skin of different people appears to vary over a wide
range, however the differ is much less in colour
(chromaticity) than brightness
– detection of skin area in digital image are more
practical and easy to implement.
5. Our Research Approach
• Skin colour is represented in 3 colour
space (rg, HSB and YCbCr)
• Using mathematical model that is
generated from face images
• Implement morphological filters for
enhancing face image
• Apply 4-neigbourhood ellipse
representation for localizing face
• Using local face databases for experiment
7. Generating Face Skin Model
• Calculate mean and covariance
chromaticity of training images for each
colour space
• Training images are prepared semi-
manually
Mk =
nX
i = 1
1
±i
Ti
Mk =
1
±1
T1 +
1
±2
T2 + ¢¢¢+
1
±n ¡ 1
Tn¡ 1 +
1
±n
Tn
10. Face Detection
Algorithm
where Pski n (i; j ) is probability of
pixel P as skin pixel if included
in distribution skin model DM k
for every colour spaces Rn .
Pski n (i; j ) = Pski n (i; j ) 2 DM k
8 P(i; j ) ^ 8 Rn
13. Concluding remarks
• Skin colour is modelled using using mean-
covariance characteristics
• Skin colour is represented in 3 colour space
• Skin model is used for face detection, with
support morphological filter dan 4-neigborhood
ellipse generation
• Experiment has been performed using 7 sets of
face database, >>3000 face images
• Accuracy needs to be improved
14. Next Work
• Multiple image detection
• Symmetry and features detection
• Adding geometric-based detection to
increase detection accuracy
• Algorithm improvement for efficient yet
faster detection
• Realtime face detection
• Face recognition module
15. Mathematical Model of Skin
Color for Face Detection
Setiawan Hadi, Adang Suwandi A,
Iping Supriana S, Farid Wazdi
Universitas Padjadjaran, Bandung, Indonesia
Institut Teknologi Bandung, Indonesia