simple-skin-detection/skin_detection.py at master · flagman/simple-skin-detection · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
import cv2
import numpy as np

# ---- START FUNCTIONS ----#

# display an image plus label and wait for key press to continue
def display_image(image, name):
    window_name = name
    cv2.namedWindow(window_name)
    cv2.imshow(window_name, image)
    cv2.waitKey()
    cv2.destroyAllWindows()


# segment using otsu binarization and thresholding
def segment_otsu(image_grayscale, img_BGR):
    threshold_value, threshold_image = cv2.threshold(image_grayscale, 0, 255,cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)
    display_image(threshold_image, "otsu")
    threshold_image_binary = 1- threshold_image/255
    threshold_image_binary = np.repeat(threshold_image_binary[:, :, np.newaxis], 3, axis=2)
    img_face_only = np.multiply(threshold_image_binary, img_BGR)
    return img_face_only

# ---- MAIN ----#

# read in image into openCV BGR and grayscale
image_path = "images/img_1.jpg"

img_BGR = cv2.imread(image_path, 3)
display_image(img_BGR, "BGR")

img_grayscale = cv2.cvtColor(img_BGR, cv2.COLOR_BGR2GRAY)
display_image(img_grayscale, "grayscale")

# foreground and background segmentation (otsu)
img_face_only = segment_otsu(img_grayscale, img_BGR)
display_image(img_face_only, "segmented BGR")

# convert to HSV and YCrCb color spaces and detect potential pixels
img_HSV = cv2.cvtColor(img_face_only.astype(np.uint8), cv2.COLOR_BGR2HSV)
img_YCrCb = cv2.cvtColor(img_face_only.astype(np.uint8), cv2.COLOR_BGR2YCrCb)

display_image(img_HSV, "HSV")
display_image(img_YCrCb, "YCrCb")

# aggregate skin pixels
blue = []
green = []
red = []

height, width, channels = img_face_only.shape

for i in range (height):
    for j in range (width):
        if((img_HSV.item(i, j, 0) <= 170) and (140 <= img_YCrCb.item(i, j, 1) <= 170) and (90 <= img_YCrCb.item(i, j, 2) <= 120)):
            blue.append(img_face_only[i, j].item(0))
            green.append(img_face_only[i, j].item(1))
            red.append(img_face_only[i, j].item(2))
        else:
            img_face_only[i, j] = [0, 0, 0]

display_image(img_face_only, "final segmentation")

# determine mean skin tone estimate
skin_tone_estimate_BGR = [np.mean(blue), np.mean(green), np.mean(red)]
print("mean skin tone estimate (BGR)", skin_tone_estimate_BGR[0], skin_tone_estimate_BGR[1], skin_tone_estimate_BGR[2], "]")