IEEE Consumer Electronics Magazine - March 2018 - 66

region where skin color gathers. The ellipse models are given
in (2) and (3), where a = 25.39, b = 14.03, ec x = 1.60,
ec y = 2.41, c x = 109.38, and c y = 152.05:

With the rapid development
of consumer electronics, VR is
attracting a great deal of
research attention.
Skin color has different distribution in various color spaces,
so a good color space can make a great contribution to the
accuracy of hand detection. We choose YCbCr because skin
colors gather in an ellipse region, which is easy to detect. In
YCbCr space, luminance and chromaticity are independent,
and Y indicates luminance of colors and Cb and Cr correspond
to the chromaticity of blue and red separately. This reduces the
redundancy present in RGB color channels and represents the
color with statistically independent components. YCbCr is one
of the most popular choices for skin detection [34].
Images caught by cameras are in RGB space, but they can
be transformed into YCbCr through (1), which is a pixelbased calculation:
Y
0.257 0.504 0.098 R
16
Cb
=
0
.
148
0
.
291
0
.
439
G
+
128
> H >
H > H > H.
Cr
0.439 - 0.368 - 0.071 B
128

(1)

Because the mapping is linear with RGB information,
luminance Y cannot be totally independent on chromaticity.
This results in the skin region changing nonlinearly with
luminance. To avoid the influence of luminance, a nonlinear
transform of YCbCr is carried out, and then we get an ellipse

(a)

(b)

(c)

(d)

(x - ec x) 2 (y - ec y) 2
+
= 1,
a2
b2

(2)

x
cos i sin i Cb - c x
; E=;
E;
E.
y
- sin i - cos i Cr - c y

(3)

Whether a pixel belongs to a hand or not is up to whether
its color is located in the ellipse region. If it is, the pixel is set
to be 1, and otherwise it is 0, such that we can roughly detect
the hand in images.
However, it should be noted that something in the background with color similar to skin color can cause noise, which
is introduced in the hand region and identified by color detection (seen in Figure 1). Under the premise that there are no
large skin-colored objects other than hands in images, we can
choose the region with the largest area as the target hand
region. Noise unconnected with the hand region can then be
removed. In terms of the noise connected with the hand, a morphological process can be used to remove it (seen in Figure 2).
First, erosion and dilation are carried out to make the region
smooth, and then filtering is done to remove disturbing noise.
Finally, holes in the region are filled to make a complete hand.
Then, we get binary images with hand regions labeled 1.

GESTURE RECOGNITION
Chinese characters are combinations of strokes. When a user
writes characters, his or her hands move in some particular
order. Some movement corresponds to strokes, and others are
just the transition between strokes. We define two gestures to
represent these two types of hand movements (seen in Figure 3).
The valid gesture, a hand with the index finger raised up, indicates a user is writing strokes. The invalid gesture, a closed
hand, indicates a hand is moving between strokes.

FIGURE 1. The hand segment and feature extraction, with (a) and
(b) showing skin color detection and (c) and (d) showing the morphological process.

Binary
Defect
Image

Erosion
and
Dilation

Hole
Filling

Filtering

FIGURE 2. The morphological process.
66 IEEE Consumer Electronics Magazine

^

MARCH 2018

Repaired
Image

(a)

(b)

FIGURE 3. The gesture definition: (a) a valid gesture used when a
user writes strokes and (b) an invalid gesture used when a user's
hands move between strokes.
Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - March 2018
