def resize_pad():
srcdir0 = '/home/kevin/PycharmProjects/autograde/dataset/svm/data-0810'
# srcdir0 = '/home/kevin/PycharmProjects/autograde/dataset/svm/data-0810-aug'
dstdir0 = hp.mkdir(
'/home/kevin/PycharmProjects/autograde/dataset/svm/data-0810-28')
# for cls in range(22):
for cls in [22]:
srcdir = os.path.join(srcdir0, str(cls))
dstdir = hp.mkdir(os.path.join(dstdir0, str(cls)))
file_list = os.listdir(srcdir)
print(len(file_list))
for file in file_list:
if file.endswith('.jpg') or file.endswith('.png'):
src_full = os.path.join(srcdir, file)
dst_full = os.path.join(dstdir, file[:-4] + '.jpg')
print(dst_full)
gray = cv2.imread(src_full, 0)
gray = imu.strip_white_boder(gray)
result = adjust_img(gray)
cv2.imwrite(dst_full, result)
def predict_with_proba(self, gray, var_thres=0.01):
im = imu.strip_white_boder(gray)
im = adjust_img(im)
sample = im.ravel() / 255.0
var = np.var(self.model.predict_proba(sample.reshape(1, -1)))
if var > var_thres:
label = self.model.predict(sample.reshape(1, -1))
else:
label = [22]
return label
def check_position(img, p1, p2):
"""Check two positions of characters.
Args:
img: A gray image object.
p1: The position of number character.
p2: The position of down or dot character.
Returns:
A boolean indicates if the positions of characters is reasonable.
"""
img1 = img[p1[1]:p1[3], p1[0]:p1[2]]
img2 = img[p2[1]:p2[3], p2[0]:p2[2]]
im = imu.strip_white_boder(img1)
t1, b1 = get_vertical_white_edge_dis(img1)
t2, b2 = get_vertical_white_edge_dis(img2)
if b2 - b1 > im.shape[0] / 4 or t2 - t1 < im.shape[0] / 2 or p2[0] - p1[
2] > im.shape[1]:
return False
return True
def all_strip_white():
categories = range(22) #不考虑其他类
for c in categories:
srcdir = os.path.join('../dataset/svm/data-0807', str(c))
outdir = os.path.join('../dataset/svm/data-0810', str(c))
if not os.path.exists(outdir):
os.mkdir(outdir)
papers = os.listdir(srcdir)
# papers = ['120190703153331060.jpg']
for paper in papers:
if paper.endswith('.jpg') or paper.endswith('.png'):
imgfile = os.path.join(srcdir, paper)
outfile = os.path.join(outdir, paper)
im = cv2.imread(imgfile, 0)
imout = imu.strip_white_boder(im)
cv2.imwrite(outfile, imout)
def augment(gray, shift_low=-10):
angle = (2 * np.random.random() - 1) * 5.0
# print(angle)
rotated = rotate_image(gray, angle)
# imu.imshow_(rotated)
rotated = imu.strip_white_boder(rotated)
# imu.imshow_(rotated)
shift = np.random.randint(low=shift_low, high=50)
gray = rotated
if shift >= 0:
gray = cv2.add(gray, shift)
else:
bw = imu.preprocess_bw_inv(gray)
gray[bw > 0] = gray[bw > 0] + shift
# print(shift)
# if shift>0:
# gray = cv2.add(gray, shift)
# else:
# gray = cv2.subtract(gray, shift)
# imu.imshow_(gray)
h, w = gray.shape
hcut = min(2, h // 5)
wcut = min(2, w // 5)
top = np.random.randint(hcut + 1)
bottom = h - np.random.randint(hcut + 1)
left = np.random.randint(wcut + 1)
right = w - np.random.randint(wcut + 1)
return gray[top:bottom, left:right]
def predict_proba(self, gray):
im = imu.strip_white_boder(gray)
im = adjust_img(im)
sample = im.ravel() / 255.0
return self.model.predict_proba(sample.reshape(1, -1))