Age and Gender Estimation Using Shifting and Re-scaling of Local Regions - Technologies and Applications of Artificial Intelligence - page 9

sing Fig. 6. The accuracy of gend

different shift amounts. The be

der classification from the voting of the different regions u
est result is at 90 pixels shift, 91.05%.

5.2 The Effect of Scalin ng on Gender Estimation

In this section, we will show
ing table (5-2), the first co
while the second column s
original size and 1.2 of its o
result with scaling to 0.5 o
along with no scaling.

w the added benefit of using multiple scales. In the follo
olumn shows the results without any scaling done. Me
shows the results when each patch is scaled to 0.8 of
original size along with no scaling. The third bar shows
of the original patch size and 1.5 of the original patch s

ow-
ean-
f its
the
size

Table 3. The accuracy of th
different scales and a shift of 9

he gender classification on different regions separately u
90 pixels

sing

[1.0] [0.8, 1.0, 1.2] [0.5 ,1.0, 1.5]

Left Eyebrow
Right Eyebrow
Left Eye
Right Eye
Nose
Mouth
Face (60 pixel shift)

81.53% 84.67% 83.44%

84.09% 85.68% 85.09%

87.75% 88.73%

89.03

86.47% 88.61%

88.59%

85.59% 85.94% 86.54%

87.53% 88.93% 89.51%

88.42% 89.49% 88.72%

Voting 91.05% 92.88% 92.78%