def __merge(self, img_even, img_odd, row_col):
if row_col == 0:
img_even = cv2.transpose(img_even)
img_odd = cv2.transpose(img_odd)
img = array([eo for e in zip(img_even, img_odd) for eo in e]) # merge two matrix
if row_col == 0:
img = cv2.transpose(img)
return img
def __split(self, origin_img, row_col):
img = deepcopy(origin_img)
if row_col == 0:
img = cv2.transpose(img)
if len(img) % 2 == 1: # number of lines is odd
img = r_[img, [img[-1]]] # copy the last line
img_even = img[0::2]
img_odd = img[1::2]
if row_col == 0:
img_even = cv2.transpose(img_even)
img_odd = cv2.transpose(img_odd)
return img_even, img_odd
def _cv2_trim(self) -> bool:
"""
Remove black borders from a cv2 image array.
This method is a f*****g waste of time as most sources are already
properly cropped. We need to use it because of a few shitty WEB sources.
F*****g unbelievable.
:param cv2_image: cv2 image array
"""
logger.info("Trying to remove black borders with cv2")
og_w, og_h = self._cv2.shape[1], self._cv2.shape[0]
logger.debug("Original dimensions: %dx%d", og_w, og_h)
og_quotient = og_w / og_h
first_img = _remove_lateral_cv2(self._cv2)
tmp_img = cv2.transpose(first_img)
tmp_img = cv2.flip(tmp_img, flipCode=1)
if tmp_img is None:
raise exceptions.InvalidRequest("Possible all-black image found")
final = _remove_lateral_cv2(tmp_img)
out = cv2.transpose(final)
final_img = cv2.flip(out, flipCode=0)
if final_img is None:
raise exceptions.InvalidRequest("Possible all-black image found")
new_w, new_h = final_img.shape[1], final_img.shape[0]
logger.debug("New dimensions: %dx%d", new_w, new_h)
new_quotient = new_w / new_h
if abs(new_quotient - og_quotient) > 0.9:
logger.info("Possible bad quotient found: %s -> %s", og_quotient,
new_quotient)
return False
width_percent = (100 / og_w) * new_w
height_percent = (100 / og_h) * new_h
if any(percent <= 65 for percent in (width_percent, height_percent)):
logger.info("Possible bad trim found: %s -> %s", width_percent,
height_percent)
return False
self._cv2 = final_img
return True
def _cv2_trim(self) -> bool:
"""
Remove black borders from a cv2 image array.
:param cv2_image: cv2 image array
"""
logger.info("Trying to remove black borders with cv2")
og_w, og_h = self.cv2.shape[1], self.cv2.shape[0]
logger.debug("Original dimensions: %dx%d", og_w, og_h)
og_quotient = og_w / og_h
first_img = _remove_lateral_cv2(self.cv2)
tmp_img = cv2.transpose(first_img)
tmp_img = cv2.flip(tmp_img, flipCode=1)
if tmp_img is None:
raise exceptions.NothingFound("Possible all-black image found")
final = _remove_lateral_cv2(tmp_img)
out = cv2.transpose(final)
final_img = cv2.flip(out, flipCode=0)
new_w, new_h = final_img.shape[1], final_img.shape[0]
logger.debug("New dimensions: %dx%d", new_w, new_h)
new_quotient = new_w / new_h
if abs(new_quotient - og_quotient) > 0.9:
logger.info("Possible bad quotient found: %s -> %s", og_quotient,
new_quotient)
return False
width_percent = (100 / og_w) * new_w
height_percent = (100 / og_h) * new_h
if any(percent <= 65 for percent in (width_percent, height_percent)):
logger.info("Possible bad trim found: %s -> %s", width_percent,
height_percent)
return False
self.cv2 = final_img
return True
def MyHitMiss(imgSrc: np.ndarray, hit: np.ndarray,
miss: np.ndarray) -> np.ndarray:
imgDst1 = Erode(imgSrc, hit)
imgDst2 = Not(Dilate(imgSrc, cv.transpose(miss)))
imgDst = np.empty(imgSrc.shape, dtype=np.uint8)
h, w = imgSrc.shape
for (i, L1, L2) in zip(range(h), imgDst1, imgDst2):
for (j, item1, item2) in zip(range(w), L1, L2):
imgDst[i, j] = min(item1, item2)
return imgDst
def preprocess(img, imgSize, dataAugmentation=False):
"put img into target img of size imgSize, transpose for TF and normalize gray-values"
# there are damaged files in IAM dataset - just use black image instead
if img is None:
img = np.zeros([imgSize[1], imgSize[0]])
# increase dataset size by applying random stretches to the images
if dataAugmentation:
stretch = (random.random() - 0.5) # -0.5 .. +0.5
wStretched = max(int(img.shape[1] * (1 + stretch)),
1) # random width, but at least 1
img = cv2.resize(
img, (wStretched,
img.shape[0])) # stretch horizontally by factor 0.5 .. 1.5
# create target image and copy sample image into it
(wt, ht) = imgSize
(h, w) = img.shape
fx = w / wt
fy = h / ht
f = max(fx, fy)
newSize = (max(min(wt, int(w / f)), 1), max(
min(ht, int(h / f)),
1)) # scale according to f (result at least 1 and at most wt or ht)
img = cv2.resize(img, newSize)
target = np.ones([ht, wt]) * 255
target[0:newSize[1], 0:newSize[0]] = img
# transpose for TF
img = cv2.transpose(target)
# normalize
(m, s) = cv2.meanStdDev(img)
m = m[0][0]
s = s[0][0]
img = img - m
img = img / s if s > 0 else img
return img
