Designing an AI-Based Virtual Try-On Web Application

. 2022 May 18;22(10):3832. doi: 10.3390/s22103832

Algorithm 1:Face size estimation algorithm.

Input: the face image, the key-points detected on the face, and the estimated distance for the ears in pixels.

Output: the estimated face size as the ear-to-ear distance in millimeters.

1:
functionFaceSizeEstimation( $i n p u t_i m g, k p t s, e a r 2 e a r_p i x e l s$ )
2:
$e y e_k p t s \leftarrow$ fetch_eyes_kpts $(k p t s)$ ▹ left, right eyes
3:
$p a d d i n g \leftarrow 10$
4:
$p x 2 m m \leftarrow 0$
5:
$d e t e c t i o n_c o u n t \leftarrow 0$
6:
for $e y e_k p t s$ in $e y e s_k p t s$ do
7:
$e y e_i m g \leftarrow$ crop_image $i n p u t_i m a g e, e y e_k p t s, p a d d i n g)$
8:
$e y e_i m g \leftarrow$ gamma_correction $e y e_i m g)$
9:
$e y e_i m g \leftarrow e y e_i m g [:, :, 0]$ ▹ use red channel only
10:
$m a s k \leftarrow$ drn_d_22_process $e y e_i m g)$ ▹ predicts the iris segmentation mask
11:
$i r i s \leftarrow$ detect_hough_circle $m a s k)$ ▹ fits a circle over the segmentation mask
12:
$m m \leftarrow 11.71 / (2 * i r i s . r a d i u s)$ ▹ size of a pixel in millimeters
13:
if mm is not ’nan’ then ▹ estimation successful
14:
$p x 2 m m \leftarrow p x 2 m m + m m$
15:
$d e t e c t i o n_c o u n t \leftarrow d e t e c t i o n_c o u n t + 1$
16:
if $d e t e c t i o n_c o u n t > 0$ then
17:
$p x 2 m m \leftarrow p x 2 m m / d e t e c t i o n_c o u n t$ ▹ average size of a pixel in millimeters
18:
else
19:
raise Exception(’Face size estimation failed!’)
20:
return $e a r 2 e a r_p i x e l s * p x 2 m m$ ▹ ear to ear distance in millimeters