def mask_image(self, img):
#blur_k_size = (11, 11)
#img = cv2.blur(img, blur_k_size)
w, h, c = img.shape
skin_mask = sd.mask_skin(img)
# determine candidates for classification using mask
skin_mask_reshaped = skin_mask.reshape(-1, 1)
candidates = (np.nonzero(skin_mask_reshaped))[0]
# reshaped image for collecting pixels
reshaped_img = img.reshape(-1, 3)
out = np.zeros(w*h, np.uint8)
# classify
start_time = time.time()
logits = self.classifier.classify(reshaped_img[candidates])
# collect skin pixels
skin_idx = np.nonzero(np.array(logits) == 1)
out[candidates[skin_idx]] = 255
out = out.reshape(w, h)
kernel= cv2.getStructuringElement(\
cv2.MORPH_ELLIPSE,(5,5))
out = cv2.erode(out, kernel, iterations = 1)
return out
def collect_data(self, img1, img2):
mask1 = sd.mask_skin(img1)
mask2 = sd.mask_skin(img2)
and_mask = cv2.bitwise_and(mask1, mask2)
# mask neg image of min_image
neg = self.collect_neg_data(img1, and_mask)
# mask pos image of min_image
pos = self.collect_pos_data(img1, mask1, and_mask)
# mask pos image of max_image
pos = np.append(pos, \
self.collect_pos_data(img2, \
mask2, and_mask), axis=0)
return pos, neg
def shake_detect(self, prev_frame, curr_frame):
h, w, c = curr_frame.shape
blur_size = 2 * round(w * 5 / 320) + 1
threshold = 0.8 * 255
copy = curr_frame
curr_frame = cv2.medianBlur(curr_frame, blur_size)
prev_frame = cv2.medianBlur(prev_frame, blur_size)
mask = sd.mask_skin(curr_frame)
diff = abs_diff(curr_frame, prev_frame)
diff = cv2.cvtColor(diff, cv2.COLOR_BGR2GRAY)
#diff[diff < threshold] = 0
diff = cv2.bitwise_and(diff, diff, mask=mask)
diff = erode_frame(diff)
cv2.imshow('diff', diff)
k = cv2.waitKey(5) & 0xFF
return diff, self.shaker.shake_detect(diff, copy)
if __name__ == '__main__':
# test data
folder_dir = "../test-data/"
img1_dir = folder_dir + "img_max.png"
img2_dir = folder_dir + "img_min.png"
test_dir = folder_dir + "img_test.png"
'''
img1_dir = "../test-data/classifier_test1.png"
img2_dir = "../test-data/classifier_test2.png"
test_dir = "../test-data/classifier_test3.png"
'''
min_img = cv2.imread(img1_dir)
min_img = cv2.resize(min_img, None, fx=0.1, fy=0.1, interpolation=cv2.INTER_AREA)
max_img = cv2.imread(img2_dir)
max_img = cv2.resize(max_img, None, fx=0.1, fy=0.1, interpolation=cv2.INTER_AREA)
test_img = cv2.imread(test_dir)
test_img = cv2.resize(test_img, None, fx=0.1, fy=0.1, interpolation=cv2.INTER_AREA)
copy = test_img
start_time = time.time()
scc = SkinColorClassifier(min_img, max_img)
out = scc.mask_image(test_img)
print("Whole Time duration: " + str(time.time() - start_time))
print("Test Image Size: " + str(test_img.shape))
cv2.imshow('fixed mask', sd.mask_skin(copy))
cv2.imshow('rf mask', out)
cv2.waitKey(0)
import cv2
import os
import skin_detection as sd
from random import *
i = 1
cam = cv2.VideoCapture(0)
BIG_NUM = 1000000000
while True:
ret_val, img = cam.read()
frame_size = (240, 180)
img = cv2.resize(img, frame_size)
img = sd.mask_skin(img)
cv2.imshow('my webcam', img)
k = cv2.waitKey(1)
out_dir = "../data/train-data/"
folder = out_dir + "R/"
if k == 27:
break
elif k == 49:
i = 1
print("Changed to R")
folder = out_dir + "R/"
elif k == 50:
print("Changed to P")
folder = out_dir + "P/"
elif k == 51:
ret, prev_frame = cap.read()
prev_frame = render_frame(prev_frame)
frame_cnt = 0
threshold = 0.8 * 255
yhistory1 = []
yhistory2 = []
while cap.isOpened():
ret, curr_frame = cap.read()
if ret:
curr_frame = render_frame(curr_frame)
mask = sd.mask_skin(curr_frame)
diff = frame_diff(curr_frame, prev_frame)
diff = cv2.cvtColor(diff, cv2.COLOR_BGR2GRAY)
diff[diff < threshold] = 0
diff = cv2.bitwise_and(diff, diff, mask=mask)
diff = erode_frame(diff)
diff = cv2.cvtColor(diff, cv2.COLOR_GRAY2BGR)
idx1 = np.argwhere(diff > 0)
cnt1 = len(idx1)
y_sum1 = 0
for i in idx1:
y_sum1 += i[0]
if cnt1 is not 0:
yhistory1.append(y_sum1/cnt1)
elif len(yhistory1) != 0: