def compute_checkbox_position(blank_im):
binary = convert_to_binary(255 - blank_im)
labels, n = morph.label(binary)
h, w = binary.shape
minsize = 40
# find small dash in img
sums = measurements.sum(binary, labels, range(n + 1))
sums = sums[labels]
good = minimum(binary, 1 - (sums > 0) * (sums < minsize))
junk_cc = np.bitwise_xor(good, binary)
# temporary fix: add bottom line
junk_cc[h-1:, :] = np.ones((1, w))
junk_cc = morph.r_dilation(junk_cc, (7,7))
junk_cc = morph.r_closing(junk_cc, (9,9))
# find hole using morphology
hole = morph.fill_hole(junk_cc)
hole = hole - junk_cc
# locate holes position
labels, n = morph.label(hole)
objects = morph.find_objects(labels)
objects = sorted(objects, key=lambda b: sl.center(b))
area_thres = 0.4 * (amax([sl.area(b) for b in objects]) if len(objects) > 0 else 0)
boxes = [[b[0].start, b[1].start, b[0].stop, b[1].stop] for b in objects if sl.area(b) > area_thres]
return boxes, convert_binary_to_normal_im(hole)
def pre_check_line(oriline):
if oriline.shape[0] < 10:
return False
if 1.0 * oriline.shape[1] / oriline.shape[0] < 1.28:
return False
if mean(oriline) < 0.35:
return False
if oriline.shape[0] > 25:
line = sauvola(oriline,
w=oriline.shape[0] * 3 / 4,
k=0.05,
reverse=True,
scaledown=20.0 / oriline.shape[0])
else:
line = sauvola(oriline,
w=oriline.shape[0] * 3 / 4,
k=0.05,
reverse=True)
if mean(line) < 0.15:
return False
_, n = morph.label(line)
n = 1.0 * n / oriline.shape[1] * oriline.shape[0]
if n > 15:
return False
return True
def remove_hlines(self, binary, scale, maxsize=10):
labels, _ = morph.label(binary)
objects = morph.find_objects(labels)
for i, b in enumerate(objects):
if sl.width(b) > maxsize * scale:
labels[b][labels[b] == i + 1] = 0
return np.array(labels != 0, 'B')
def remove_hlines(binary,scale,maxsize=10):
labels,_ = morph.label(binary)
objects = morph.find_objects(labels)
for i,b in enumerate(objects):
if sl.width(b)>maxsize*scale:
labels[b][labels[b]==i+1] = 0
return array(labels!=0,'B')
def firstAnalyse(binary):
binaryary = morph.r_closing(binary.astype(bool), (1,1))
labels,_ = morph.label(binaryary)
objects = morph.find_objects(labels) ### <<<==== objects here
bysize = sorted(range(len(objects)), key=lambda k: sl.area(objects[k]))
# bysize = sorted(objects,key=sl.area)
scalemap = zeros(binaryary.shape)
smalldot = zeros(binaryary.shape, dtype=binary.dtype)
for i in bysize:
o = objects[i]
if amax(scalemap[o])>0:
# mask = where(labels[o] != (i+1),uint8(255),uint8(0))
# binary[o] = cv2.bitwise_and(binary[o],binary[o],mask=mask)
continue
scalemap[o] = sl.area(o)**0.5
scale = median(scalemap[(scalemap>3)&(scalemap<100)]) ### <<<==== scale here
for i,o in enumerate(objects):
if (sl.width(o) < scale/2) or (sl.height(o) < scale/2):
smalldot[o] = binary[o]
if sl.dim0(o) > 3*scale:
mask = where(labels[o] != (i+1),uint8(255),uint8(0))
binary[o] = cv2.bitwise_and(binary[o],binary[o],mask=mask)
continue
return objects, smalldot, scale
def compute_line_seeds(binary,bottom,top,colseps,scale,threshold=0.2,vscale=1.0):
"""Base on gradient maps, computes candidates for baselines
and xheights. Then, it marks the regions between the two
as a line seed."""
t = threshold
vrange = int(vscale*scale)
bmarked = maximum_filter(bottom==maximum_filter(bottom,(vrange,0)),(2,2))
bmarked *= (bottom>t*amax(bottom)*t)*(1-colseps)
tmarked = maximum_filter(top==maximum_filter(top,(vrange,0)),(2,2))
tmarked *= (top>t*amax(top)*t/2)*(1-colseps)
tmarked = maximum_filter(tmarked,(1,20))
seeds = zeros(binary.shape,'i')
delta = max(3,int(scale/2))
for x in range(bmarked.shape[1]):
transitions = sorted([(y,1) for y in find(bmarked[:,x])]+[(y,0) for y in find(tmarked[:,x])])[::-1]
transitions += [(0,0)]
for l in range(len(transitions)-1):
y0,s0 = transitions[l]
if s0==0: continue
seeds[y0-delta:y0,x] = 1
y1,s1 = transitions[l+1]
if s1==0 and (y0-y1)<5*scale: seeds[y1:y0,x] = 1
seeds = maximum_filter(seeds,(1,int(1+scale)))
seeds *= (1-colseps)
#DSAVE("lineseeds",[seeds,0.3*tmarked+0.7*bmarked,binary])
seeds,_ = morph.label(seeds)
return seeds
def extract_region_from_image_cc(im, rect):
pad = 10
y0, x0, y1, x1 = rect
local_im = im[y0-pad:y1+pad, x0-pad:x1+pad].copy()
h, w = local_im.shape
box = (slice(pad, h - pad, None), slice(pad, w - pad, None))
binary = convert_to_binary(255 - local_im)
labels, _ = morph.label(binary)
objects = morph.find_objects(labels)
region_objs = []
for i, b in enumerate(objects):
if 1.0 * sl.xoverlap(b, box) * sl.yoverlap(b, box) / sl.area(b) > 0.55:
region_objs.append(b)
else:
binary[b][labels[b]==i+1] = 0
x2 = min([obj[1].start for obj in region_objs] + [pad])
x3 = max([obj[1].stop for obj in region_objs] + [w - pad])
y2 = min([obj[0].start for obj in region_objs] + [pad])
y3 = max([obj[0].stop for obj in region_objs] + [h - pad])
return convert_binary_to_normal_im(binary[y2:y3, x2:x3]), (pad-x2, x3-w+pad, pad-y2, y3-h+pad)
def compute_line_seeds(binary, bottom, top, colseps, scale, threshold, vscale):
"""Base on gradient maps, computes candidates for baselines
and xheights. Then, it marks the regions between the two
as a line seed."""
t = threshold
vrange = int(vscale * scale)
bmarked = maximum_filter(bottom == maximum_filter(bottom, (vrange, 0)),
(2, 2))
bmarked = bmarked * (bottom > t * np.amax(bottom) * t) * (1 - colseps)
tmarked = maximum_filter(top == maximum_filter(top, (vrange, 0)), (2, 2))
tmarked = tmarked * (top > t * np.amax(top) * t / 2) * (1 - colseps)
tmarked = maximum_filter(tmarked, (1, 20))
seeds = np.zeros(binary.shape, 'i')
delta = max(3, int(scale / 2))
for x in range(bmarked.shape[1]):
transitions = sorted([(y, 1) for y in find(bmarked[:, x])] +
[(y, 0) for y in find(tmarked[:, x])])[::-1]
transitions += [(0, 0)]
for l in range(len(transitions) - 1):
y0, s0 = transitions[l]
if s0 == 0: continue
seeds[y0 - delta:y0, x] = 1
y1, s1 = transitions[l + 1]
if s1 == 0 and (y0 - y1) < 5 * scale: seeds[y1:y0, x] = 1
seeds = maximum_filter(seeds, (1, int(1 + scale)))
seeds = seeds * (1 - colseps)
DSAVE("lineseeds", [seeds, 0.3 * tmarked + 0.7 * bmarked, binary])
seeds, _ = morph.label(seeds)
return seeds
def calc_line(oriline):
line = sauvola(oriline, w=oriline.shape[0] / 2, k=0.05, reverse=True)
oridense = '{:3.3f}'.format(mean(oriline))
dens = '{:3.3f}'.format(mean(line))
# rati = '{:3.3f}'.format(1.0*line.shape[1]/line.shape[0])
_, n = morph.label(line)
n = '{:3.3f}'.format(1.0 * n / oriline.shape[1] * oriline.shape[0])
return dens + '_' + n + '_' + oridense, line
def remove_vlines(binary, gray, scale, maxsize=10):
labels, _ = morph.label(binary)
objects = morph.find_objects(labels)
for i, b in enumerate(objects):
if (sl.width(b) <= 20 and sl.height(b) > 200) or (sl.width(b) <= 45 and sl.height(b) > 500):
gray[b][labels[b] == i + 1] = 140
# gray[:,b[1].start:b[1].stop]=140
labels[b][labels[b] == i + 1] = 0
return array(labels != 0, "B")
def remove_noise(line,minsize=8):
"""Remove small pixels from an image."""
if minsize==0: return line
bin = (line>0.5*amax(line))
labels,n = morph.label(bin)
sums = measurements.sum(bin,labels,range(n+1))
sums = sums[labels]
good = minimum(bin,1-(sums>0)*(sums<minsize))
return good
def text_line_segmentation_NN(image,
scale,
mask=None,
use_binary=False,
debug_image=None,
offset=(0, 0)):
h, w = image.shape[:2]
if debug_image is None:
debug_image = image
if len(debug_image.shape) < 3:
debug_image = cv2.cvtColor(image, cv2.COLOR_GRAY2BGR)
if len(image.shape) > 2:
image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
if use_binary:
image = morph.thresh_sauvola(image, k=0.15) * 255
if mask is not None:
image = (image + mask * 255).astype('uint8')
labels, _ = morph.label(image == 0)
objects = morph.find_objects(labels)
height_map = [
sl.height(o) for o in objects if sl.height(o) > 6
and sl.height(o) < 100 and sl.aspect_normalized(o) < 8
]
avg_h = max(np.nan_to_num(np.mean(height_map)), scale * 0.6)
block = Block(image, avg_h)
words = block.getWordBoundingBoxes()
lines = filter_and_merge_overlap_boxes(words,
max(avg_h, scale * 1.2) * 0.3)
lines = filter_and_merge_overlap_boxes(lines,
max(avg_h, scale * 1.2) * 0.3,
use_merge_same_line_only=True)
offset_x, offset_y = offset
# filter line by size
lines = [
l for l in lines if l[3] - l[1] > avg_h * 0.3 and l[3] - l[1] < avg_h *
2.5 and l[2] - l[0] > avg_h * 0.5
]
lines = [sl.pad_box(l, 0, (h, w)) for l in lines]
lines = [[
l[0] + offset_x, l[1] + offset_y, l[2] + offset_x, l[3] + offset_y
] for l in lines]
debug_image = block.paint(None)
return lines, debug_image
def remove_vlines(binary,gray,scale,maxsize=10):
labels,_ = morph.label(binary)
objects = morph.find_objects(labels)
for i,b in enumerate(objects):
if (sl.width(b)<=20 and sl.height(b)>200) or (sl.width(b)<=45 and sl.height(b)>500):
gray[b][labels[b]==i+1] = 140
#gray[:,b[1].start:b[1].stop]=140
labels[b][labels[b]==i+1] = 0
return array(labels!=0, 'B')
def calc(self, objects, scale):
if self.binpage is None:
return
tt = time()
bottom, top, boxmap = compute_gradmaps(self.binpage, scale)
# DSHOW('hihi', [0.5*bottom+0.5*top,self.binpage, boxmap])
seeds0 = compute_line_seeds(self.binpage, bottom, top, scale)
seeds, _ = morph.label(seeds0)
llabels = morph.propagate_labels(boxmap, seeds, conflict=0)
spread = spread_labels(seeds, maxdist=scale)
llabels = where(llabels > 0, llabels, spread * self.binpage)
segmentation = llabels * self.binpage
self.binpage = ocrolib.remove_noise(self.binpage, args.noise)
lines = psegutils.compute_lines(segmentation, scale)
binpage_reversed = 1 - self.binpage
# print 'pre line ', time() - tt
tt = time()
self.lines = []
for i, l in enumerate(lines):
tt = time()
binline = psegutils.extract_masked(
binpage_reversed, l, pad=args.pad,
expand=args.expand) # black text
binline = (1 - binline)
le = lineest.CenterNormalizer(binline.shape[0]) # white text
binline = binline.astype(float)
le.measure(binline)
binline = le.normalize(binline)
binline = where(binline > 0.5, 0, 1) # black text
# print 'line time ', time()-tt
print '-----------------------'
pilimg = Image.fromarray((binline * 255).astype(uint8))
pred_legacy = pytesseract.image_to_string(pilimg,
lang='eng',
config='--oem 0 --psm 7')
print '00', pred_legacy
pred_lstm = pytesseract.image_to_string(pilimg,
lang='eng',
config='--oem 1 --psm 7')
print '11', pred_lstm
# ASHOW('line',binline, scale=2.0)
## pred_both = pytesseract.image_to_string(pilimg,lang='vie', config='--oem 2 --psm 7')
## print '22', pred_both
result = psegutils.record(bounds=l.bounds,
text1=pred_legacy,
text2=pred_lstm,
img=binline)
self.lines.append(result)
def simplefirstAnalyse(binary):
binaryary = morph.r_closing(binary.astype(bool), (1,1))
labels,_ = morph.label(binaryary)
objects = morph.find_objects(labels) ### <<<==== objects here
smalldot = zeros(binaryary.shape, dtype=binary.dtype)
scale = int(binary.shape[0]*0.7)
for i,o in enumerate(objects):
if (sl.width(o) < scale/2) or (sl.height(o) < scale/2):
smalldot[o] = binary[o]
if sl.dim0(o) > 3*scale:
mask = where(labels[o] != (i+1),uint8(255),uint8(0))
binary[o] = cv2.bitwise_and(binary[o],binary[o],mask=mask)
continue
return objects, smalldot, scale
def remove_small_noise(binary, minsize = 50):
labels, n = morph.label(binary)
h, w = binary.shape
objects = morph.find_objects(labels)
space_to_edge = 10
sums = measurements.sum(binary, labels, range(n + 1))
sums = sums[labels]
good = minimum(binary, 1 - (sums > 0) * (sums < minsize))
for i, b in enumerate(objects):
cy, cx = sl.center(b)
# if component is small and close to edge
if (sl.area(b) < minsize * 1.2 and ((cx < space_to_edge or cx > w - space_to_edge) or (cy < space_to_edge or cy > h - space_to_edge))):
good[b][labels[b] == i+1] = 0
return good
def compute_boxmap(binary, scale, oriimg, threshold=(.5, 4), dtype='i'):
labels, n = morph.label(binary)
objects = morph.find_objects(labels)
boxmap = zeros(binary.shape, dtype)
for i, o in enumerate(objects):
h = sl.dim0(o)
w = sl.dim1(o)
ratio = float(h) / w if h > w else float(w) / h
if h > 2 * scale or h < scale / 3:
continue
if ratio > 8: continue
# if sl.area(o)**.5<threshold[0]*scale: continue
# if sl.area(o)**.5>threshold[1]*scale: continue
boxmap[o] = 1
return boxmap
def filter_junk_cc(binary, scale, maxsize):
junk_cc = np.zeros(binary.shape, dtype='B')
text_like = np.zeros(binary.shape, dtype='B')
labels, _ = morph.label(binary)
objects = morph.find_objects(labels)
for i, b in enumerate(objects):
if sl.width(b) > maxsize * scale or sl.area(b) > scale * scale * 8 or \
sl.aspect_normalized(b) > 8 or sl.min_dim(b) < scale * 0.35:
junk_cc[b][labels[b] == i + 1] = 1
else:
if sl.width(b) > 0.3 * scale and sl.height(b) > 0.3 * scale:
text_like[b][labels[b] == i + 1] = 1
return junk_cc, text_like
def calc_typo_metric(binline):
labels,n = morph.label(binline)
objects = morph.find_objects(labels)
filtered = []
max_h = 0
for o in objects:
h = sl.dim0(o)
w = sl.dim1(o)
if h > binline.shape[0]*0.98: continue
if h < 3 or w < 3: continue
if (h > binline.shape[0]*0.2 and w > binline.shape[0]*0.2) or \
(o[0].start > binline.shape[0]/2 and o[1].stop > binline.shape[1]/4 and o[1].stop < 3*binline.shape[1]/4 and o[0].stop < binline.shape[0]*0.98):
filtered.append(o)
if h > max_h:
max_h = h
filtered.sort(key=lambda x:x[1].start)
prech = None
zoomforsee = 4
infoheight=50
info = np.zeros((infoheight*2+binline.shape[0]*zoomforsee,binline.shape[1]*zoomforsee))
for ch in filtered:
h = sl.dim0(ch)
w = sl.dim1(ch)
if prech is not None and ch[1].start < (prech[1].start + prech[1].stop)/2: continue
cv2.putText(info,'{:3.2f}'.format(1.0*w/max_h),\
((ch[1].start)*zoomforsee, int(infoheight*0.4)), cv2.FONT_HERSHEY_SIMPLEX, \
0.5,1.0,1)
if prech is None:
cv2.putText(info,'{:3d}'.format(max_h),\
((ch[1].start)*zoomforsee, int(infoheight*0.9)), cv2.FONT_HERSHEY_SIMPLEX, \
0.5,1.0,1)
else:
space = ch[1].start - prech[1].stop
dist = ch[1].stop - prech[1].stop
cv2.putText(info,'{:3.2f}'.format(1.0*space/max_h),\
((prech[1].stop)*zoomforsee, int(infoheight*0.9)), cv2.FONT_HERSHEY_SIMPLEX, \
0.5,1.0,1)
cv2.putText(info,'({:3.2f})'.format(1.0*dist/max_h),\
((prech[1].stop)*zoomforsee, int(infoheight*1.4)), cv2.FONT_HERSHEY_SIMPLEX, \
0.5,1.0,1)
prech = ch
info[infoheight*2:,:] = cv2.resize(binline, (zoomforsee*binline.shape[1], zoomforsee*binline.shape[0]))
return (info*250).astype(np.uint8), filtered
def compute_line_seeds(binary, bottom, top, colseps, scale):
"""Base on gradient maps, computes candidates for baselines
and xheights. Then, it marks the regions between the two
as a line seed."""
t = args.threshold # 0.2###############################################################more focus here for bigger fonts.!!!
# print "SbU", t
vrange = int(args.vscale * scale)
bmarked = maximum_filter(bottom == maximum_filter(bottom, (vrange, 0)),
(2, 2))
bmarked *= array((bottom > t * amax(bottom) * t) * (1 - colseps),
dtype=bool)
tmarked = maximum_filter(top == maximum_filter(top, (vrange, 0)), (2, 2))
tmarked *= array((top > t * amax(top) * t / 2) * (1 - colseps), dtype=bool)
tmarked = maximum_filter(tmarked, (1, 20))
testseeds = zeros(binary.shape, 'i')
seeds = zeros(binary.shape, 'i')
delta = max(3, int(scale / 2))
for x in range(bmarked.shape[1]):
transitions = sorted([(y, 1) for y in psegutils.find(bmarked[:, x])] +
[(y, 0)
for y in psegutils.find(tmarked[:, x])])[::-1]
transitions += [(0, 0)]
for l in range(len(transitions) - 1):
y0, s0 = transitions[l]
if s0 == 0:
continue
seeds[y0 - delta:y0, x] = 1
y1, s1 = transitions[l + 1]
if s1 == 0 and (y0 - y1) < 5 * scale:
seeds[y1:y0, x] = 1
seeds = maximum_filter(seeds, (1, int(1 + scale)))
seeds *= (1 - colseps)
###
####imsave('/home/gupta/Documents/seeds.png', seeds)
####imsave('/home/gupta/Documents/top_bottom.png', [testseeds,0.3*tmarked+0.7*bmarked,binary])
###
####imsave('/home/gupta/Documents/lineseeds.png', [seeds,0.3*tmarked+0.7*bmarked,binary])
# DSAVE("lineseeds",[seeds,0.3*tmarked+0.7*bmarked,binary])
seeds, _ = morph.label(seeds)
return seeds
def compute_line_seeds(self, binary, bottom, top, colseps, scale):
"""Base on gradient maps, computes candidates for baselines
and xheights. Then, it marks the regions between the two
as a line seed."""
t = self.parameter['threshold']
vrange = int(self.parameter['vscale'] * scale)
bmarked = maximum_filter(bottom == maximum_filter(bottom, (vrange, 0)),
(2, 2))
bmarked *= np.array((bottom > t * np.amax(bottom) * t) * (1 - colseps),
dtype=bool)
tmarked = maximum_filter(top == maximum_filter(top, (vrange, 0)),
(2, 2))
tmarked *= np.array((top > t * np.amax(top) * t / 2) * (1 - colseps),
dtype=bool)
tmarked = maximum_filter(tmarked, (1, 20))
testseeds = np.zeros(binary.shape, 'i')
seeds = np.zeros(binary.shape, 'i')
delta = max(3, int(scale / 2))
for x in range(bmarked.shape[1]):
transitions = sorted([(y, 1)
for y in psegutils.find(bmarked[:, x])] +
[(y, 0)
for y in psegutils.find(tmarked[:,
x])])[::-1]
transitions += [(0, 0)]
for l in range(len(transitions) - 1):
y0, s0 = transitions[l]
if s0 == 0:
continue
seeds[y0 - delta:y0, x] = 1
y1, s1 = transitions[l + 1]
if s1 == 0 and (y0 - y1) < 5 * scale:
seeds[y1:y0, x] = 1
seeds = maximum_filter(seeds, (1, int(1 + scale)))
seeds *= (1 - colseps)
seeds, _ = morph.label(seeds)
return seeds
def cut_dash_line(im, num_cells):
binary = convert_to_binary(255-im, thres=0.5)
labels, _ = morph.label(binary)
objects = morph.find_objects(labels)
scale = int(round(1.0 * binary.shape[1] / num_cells + 0.2))
h = binary.shape[0] - 1
# list to store objects for each cell
cells = [[] for _ in range(num_cells)]
cell_ims = []
for i, b in enumerate(objects):
# only process object with width < 2 x scale
if sl.width(b) < 2 * scale:
x1, x2 = b[1].start, b[1].stop
mid_x = (x1 + x2) // 2
cell_index = np.median([x1 // scale, x2 // scale, mid_x // scale]).astype(int)
#print(cell_index)
# handle case where digit from 2 cells connected
if x2 - (cell_index + 1) * scale > 0.3 * scale:
temp_b = (b[0], slice(b[1].start, (cell_index + 1) * (scale + 1), None))
print("2 char connected!!!")
else:
temp_b = b
cells[cell_index].append(temp_b)
for i, c in enumerate(cells):
if len(c) > 0:
x1 = min([obj[1].start for obj in c])
x2 = max([obj[1].stop for obj in c])
cell_ims.append(normalize_cell_img(im[0:h, x1:x2]))
else:
blank = np.zeros((h, scale))
cell_ims.append(normalize_cell_img(convert_binary_to_normal_im(blank)))
return cell_ims
def printResult(self, outputfile):
# Some pre-process
# print 'text area before'
# cv2.imshow('patch', self.patch)
# cv2.waitKey(-1)
if self.name == 'CMND cu - 9 so':
k = 0.45
else:
k = 0.33
patch = sharpen(self.patch)
binary = sauvola(patch,
w=int(self.template.shape[1] / 24.5 * 2),
k=k,
scaledown=0.5,
reverse=True)
binary = cv2.bitwise_and(binary, binary, mask=self.patch_mask)
# print 'text area after'
# cv2.imshow('patch', binary*255)
# cv2.waitKey(-1)
dotremoved = binary
scale = self.scale
# Line extraction copied from Ocropus source code
bottom, top, boxmap = compute_gradmaps(dotremoved, scale)
seeds0 = compute_line_seeds(dotremoved, bottom, top, scale)
seeds, _ = morph.label(seeds0)
llabels = morph.propagate_labels(boxmap, seeds, conflict=0)
spread = spread_labels(seeds, maxdist=scale)
llabels = where(llabels > 0, llabels, spread * dotremoved)
segmentation = llabels * dotremoved
dotremoved = ocrolib.remove_noise(dotremoved, 8)
lines = psegutils.compute_lines(segmentation, scale / 2)
binpage_reversed = 1 - dotremoved
self.lines = []
readrs = dict.fromkeys(self.linepos1.keys(), u'')
lines = sorted(lines, key=lambda x: x.bounds[1].start)
for i, l in enumerate(lines):
# Line extraction copied from Ocropus source code
binline = psegutils.extract_masked(binpage_reversed,
l,
pad=int(scale / 2),
expand=0) # black text
binline = (1 - binline)
le = lineest.CenterNormalizer(binline.shape[0]) # white text
binline = binline.astype(float)
le.measure(binline)
binline = le.normalize(binline)
# print 'normalized'
# cv2.imshow('line', binline)
# cv2.waitKey(-1)
binline = cv2.resize(binline, None, fx=2.0, fy=2.0)
# print 'resized'
# cv2.imshow('line', binline)
# cv2.waitKey(-1)
binline = where(binline > 0.5, uint8(0), uint8(255)) # black text
# print 'black text'
# cv2.imshow('line', binline)
# cv2.waitKey(-1)
# pilimg = Image.fromarray(binline)
pos = l.bounds[0].stop
left = (l.bounds[1].start < self.template.shape[1] / 2)
# Prediction using Tesseract 4.0
if pos > self.linepos1['idNumber'][0] and pos < self.linepos1[
'idNumber'][1]: #ID, all numbers
pred = ocr(
binline,
config=
'--oem 0 --psm 7 -c tessedit_char_whitelist=0123456789')
readrs['idNumber'] += pred + u' '
elif pos > self.linepos1['dateOfBirth'][0] and pos < self.linepos1[
'dateOfBirth'][1]: # DOB, number, - , /
pred = ocr(
binline,
config=
'--oem 1 --psm 7 -c tessedit_char_whitelist=0123456789-/')
readrs['dateOfBirth'] += pred + u' '
elif left and pos > self.linepos1['Gender'][
0] and pos < self.linepos1['Gender'][1]:
pred = ocr(binline, config='--oem 1 --psm 7 -l vie')
readrs['Gender'] += pred + u' '
elif (not left) and pos > self.linepos1['Dantoc'][
0] and pos < self.linepos1['Dantoc'][1]:
pred = ocr(binline, config='--oem 1 --psm 7 -l vie')
readrs['Dantoc'] += pred + u' '
elif pos > self.linepos1['NguyenQuan'][0] and pos < self.linepos1[
'NguyenQuan'][1]:
pred = ocr(binline, config='--oem 1 --psm 7 -l vie')
readrs['NguyenQuan'] += pred + u' '
elif pos > self.linepos1['fullName'][0] and pos < self.linepos1[
'fullName'][1]:
pred = ocr(binline, config='--oem 1 --psm 7 -l vie')
readrs['fullName'] += pred + u' '
# else:
# pred = ocr(binline, config='--oem 1 --psm 7 -l vie')
# print 'unknown ', unicode2ascii(pred), 'y:', l.bounds[0], 'x:', l.bounds[1]
for k in readrs:
readrs[k] = (readrs[k].replace(u'²',
u'2').replace(u'º', u'o').replace(
u'»', u'-')).strip()
if len(readrs[k]) == 0:
readrs[k] = None
if self.name == 'CMND moi - 12 so':
readrs['type'] = 'CMND Mới - 12 Số'
elif self.name == 'Can Cuoc Cong Dan':
readrs['type'] = 'Căn Cước Công Dân'
elif self.name == 'CMND cu - 9 so':
readrs['type'] = 'CMND Cũ - 9 Số'
readrs['NgayHetHan'] = None
outputfile.write(json.dumps(readrs))
def binary_objects(binary):
labels,n = morph.label(binary)
objects = morph.find_objects(labels)
return objects
def detect_table(image, scale, maxsize=10, debug_path=None):
h, w = image.shape[:2]
if len(image.shape) > 2 and image.shape[2] >= 3:
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
else:
gray = image
# kernel = np.ones((5,5),np.uint8)
# gray = 255-cv2.morphologyEx(255-gray, cv2.MORPH_CLOSE, kernel)
binary = 1 - morph.thresh_sauvola(gray, k=0.05)
junk_cc, _ = filter_junk_cc(binary, scale, maxsize)
junk_cc = morph.r_closing(junk_cc, (5, 5))
print('calculating combine sep...')
combine_sep = compute_combine_seps(junk_cc, scale)
# using closing morphology to connect disconnected edges
close_thes = int(scale * 0.15)
closed_sep = morph.r_closing(combine_sep, (close_thes, close_thes))
if debug_path is not None:
cv2.imwrite(filename[:-4] + '_bin.png',
((1 - junk_cc) * 255).astype('uint8'))
cv2.imwrite(filename[:-4] + '_sep.png',
(closed_sep * 255).astype('uint8'))
labels, _ = morph.label(closed_sep)
objects = morph.find_objects(labels)
# result table list
boxes = []
for i, b in enumerate(objects):
if sl.width(b) > maxsize * scale or sl.area(
b) > scale * scale * 10 or (sl.aspect_normalized(b) > 6
and sl.max_dim(b) > scale * 1.5):
density = np.sum(combine_sep[b])
density = density / sl.area(b)
if (sl.area(b) > scale * scale * 10 and sl.min_dim(b) > scale * 1.0
and sl.max_dim(b) > scale * 8 and density < 0.4):
# calculate projection to determine table border
w = sl.width(b)
h = sl.height(b)
region = (labels[b] == i + 1).astype('uint8')
border_pad = max(w, h)
border_thres = scale * 2
proj_x = np.sum(region, axis=0)
proj_y = np.sum(region, axis=1)
proj_x[3:] += proj_x[:-3]
proj_y[3:] += proj_y[:-3]
sep_x = np.sort([j[0] for j in np.argwhere(proj_x > 0.75 * h)])
sep_y = np.sort([j[0] for j in np.argwhere(proj_y > 0.4 * w)])
# skip if sep count < 2
if len(sep_x) < 1 or len(sep_y) < 1: continue
border_left, border_right, border_top, border_bottom = None, None, None, None
if sep_x[0] < border_pad:
border_left = sep_x[0]
if sep_x[-1] > w - border_pad:
border_right = sep_x[-1]
if sep_y[0] < border_pad:
border_top = sep_y[0]
if sep_y[-1] > h - border_pad:
border_bottom = sep_y[-1]
# print_info(border_top, border_bottom, border_left, border_right)
if all([
j is not None for j in
[border_top, border_bottom, border_left, border_right]
]):
border_right = b[1].stop - b[1].start
boxes.append([
b[1].start + border_left, b[0].start + border_top,
b[1].start + border_right, b[0].start + border_bottom
])
# boxes.append(([b[1].start, b[0].start, b[1].stop, b[0].stop]))
return boxes
def extractLines(imgpath, param):
img_grey = ocrolib.read_image_gray(imgpath)
(h, w) = img_grey.shape[:2]
img00 = cv2.resize(img_grey[h / 4:3 * h / 4, w / 4:3 * w / 4],
None,
fx=0.5,
fy=0.5)
angle = estimate_skew_angle(img00, linspace(-5, 5, 42))
print 'goc', angle
rotM = cv2.getRotationMatrix2D((w / 2, h / 2), angle, 1)
img_grey = cv2.warpAffine(img_grey, rotM, (w, h))
h, w = img_grey.shape
img_grey = cv2.normalize(img_grey.astype(float32), None, 0.0, 0.999,
cv2.NORM_MINMAX)
binary = sauvola(img_grey,
w=param.w,
k=param.k,
scaledown=0.2,
reverse=True) ### PARAM
binary = morph.r_closing(binary.astype(bool), (args.connect, 1))
binaryary = binary[h / 4:3 * h / 4, w / 4:3 * w / 4]
binary = binary.astype(np.uint8)
labels, n = morph.label(binaryary)
objects = morph.find_objects(labels)
bysize = sorted(objects, key=sl.area)
scalemap = zeros(binaryary.shape)
for o in bysize:
if amax(scalemap[o]) > 0: continue
scalemap[o] = sl.area(o)**0.5
scale = median(scalemap[(scalemap > 3) & (scalemap < 100)])
objects = psegutils.binary_objects(binary)
boxmap = zeros(binary.shape, dtype=np.uint8)
imgwidth = binary.shape[1]
imgheight = binary.shape[0]
cellwidth = 6 * scale
cellheight = 2.5 * scale
N_x = int(round(imgwidth / cellwidth))
cellwidth = int(round(imgwidth / N_x))
N_y = int(round(imgheight / cellheight))
cellheight = int(round(imgheight / N_y))
cells_list = [{}, {}, {}, {}]
def pixel2cell2id(pixel_x, pixel_y, CELLTYPE):
dx = 0
dy = 0
if CELLTYPE == 3:
pixel_x -= cellwidth / 2
pixel_y -= cellheight / 2
dx = cellwidth / 2
dy = cellheight / 2
if CELLTYPE == 2:
pixel_x -= cellwidth / 2
dx = cellwidth / 2
if CELLTYPE == 1:
pixel_y -= cellheight / 2
dy = cellheight / 2
if pixel_x <= 0 or pixel_y <= 0: return None, None
cellcoord = (pixel_x / cellwidth, pixel_y / cellheight)
cellid = cellcoord[0] + cellcoord[1] * N_x
cellcoord = (cellcoord[0] * cellwidth + dx,
cellcoord[1] * cellheight + dy)
return cellcoord, cellid
def id2cell2pixel(cellid, x, y, CELLTYPE):
cellcoord = (cellid % N_x, cellid / N_x)
pixel_x = cellcoord[0] * cellwidth + x
pixel_y = cellcoord[1] * cellheight + y
if CELLTYPE == 3:
pixel_x += cellwidth / 2
pixel_y += cellheight / 2
return cellcoord, pixel_x, pixel_y
img_grey = (cv2.cvtColor(img_grey, cv2.COLOR_GRAY2BGR) * 255).astype(
np.uint8)
for o in objects:
h = sl.dim0(o)
w = sl.dim1(o)
ratio = float(w) / h
### Dirty cheat
if ratio > 1 and ratio < 6:
recommended_width = max(int(0.6 * (o[0].stop - o[0].start)),
int(scale * 0.6), 5)
for pos in range(o[1].start + recommended_width, o[1].stop,
recommended_width):
binary[o[0].start:o[0].stop, pos:pos + 1] = np.uint8(0)
objects = psegutils.binary_objects(binary)
for o in objects:
h = sl.dim0(o)
w = sl.dim1(o)
a = h * w
# black = float(sum(binary[o]))/a
# if sl.area(o)**.5<threshold[0]*scale: continue
# if sl.area(o)**.5>threshold[1]*scale: continue
if h > 5 * scale: continue
# if h < 0.4*scale: continue
if w > 4 * scale and (h > 2 * scale or h < 0.5 * scale): continue
if a < 0.25 * scale * scale: continue
if float(h) / w > 10: continue
ratio = float(w) / h
if ratio > 10: continue
### Add object as candidate character
pixel_x, pixel_y = (o[1].start + o[1].stop) / 2, o[0].stop
for celltype in range(4):
cellcoord, cellid = pixel2cell2id(pixel_x,
pixel_y,
CELLTYPE=celltype)
if cellcoord is None or cellid is None: continue
cellbound = slice(cellcoord[1], cellcoord[1] + cellheight,
None), slice(cellcoord[0],
cellcoord[0] + cellwidth, None)
if cellid not in cells_list[celltype]:
cells_list[celltype][cellid] = SubLineFinder(
window_size=max(3, scale / 6),
cellbound=cellbound,
initChar=o)
else:
cells_list[celltype][cellid].addChar(o)
y0 = o[0].start
y1 = o[0].stop - 3 if o[0].stop - o[0].start > 8 else o[0].start + 5
x0 = o[1].start
x1 = o[1].stop - 3 if o[1].stop - o[1].start > 8 else o[1].start + 5
boxmap[y0:y1, x0:x1] = 1
for celltype in range(4):
if celltype == 0: col = (255, 0, 0)
if celltype == 1: col = (0, 255, 0)
if celltype == 2: col = (255, 255, 0)
if celltype == 3: col = (0, 0, 255)
for cellid, subline in cells_list[celltype].iteritems():
# cv2.rectangle(img_grey, (subline.cellbound[1].start+celltype, subline.cellbound[0].start+celltype), (subline.cellbound[1].stop+celltype, subline.cellbound[0].stop+celltype), col,1)
line = subline.subline()
if line is not None:
pos1 = (int(line[0][0]), int(line[0][1]))
pos2 = (int(line[1][0]), int(line[1][1]))
# print cellid, pos1, pos2
cv2.line(img_grey, pos1, pos2, col, 1)
### illustrate/debug first round
return binary, cv2.add(img_grey, (boxmap[:, :, np.newaxis] *
np.array([0, 50, 50])).astype(np.uint8))
def remove_small_noise_and_seps(im, num_cells, minsize = 30):
im = 255 - im
binary = np.array(im>0.5*(np.amin(im)+np.amax(im))).astype('uint8') # invert
h, w = im.shape
scale = int(w / num_cells)
labels, n = morph.label(binary)
objects = morph.find_objects(labels)
# remove small noise using connected components
sums = measurements.sum(binary, labels, range(n + 1))
sums = sums[labels]
good = minimum(binary, 1 - (sums > 0) * (sums < minsize))
# calculate sep bar positions from junk cc
junk_cc = np.bitwise_xor(binary, good)
# remove long connected component (solid separators)
proj_x = np.sum(binary, axis=0)
mask_x = np.tile((proj_x > h * 0.8).astype('B'), h)
solid_sep_pos = [j[0] for j in np.argwhere(proj_x > h * 0.6)]
good[mask_x] = 0
'''for i, b in enumerate(objects):
if sl.width(b) < 6 and sl.height(b) > h * 0.9:
good[b][labels[b] == i + 1] = 0
'''
if np.sum(junk_cc) > 140:
# only detect sep bars if has enough pixels
proj_x = np.sum(junk_cc, axis=0)
mask_x = proj_x > np.amax(proj_x) * 0.2
sep_pos = np.array([i[0] for i in np.argwhere(mask_x)])
start = [True] + [True if abs(sep_pos[i] - sep_pos[i-1] - scale) < 5 or abs(sep_pos[i] - sep_pos[i-1] - 2 * scale) < 5 else False for i in range(1,len(sep_pos))]
else:
sep_pos = []
if len(sep_pos) > 0:
start_sep_pos = sep_pos[start]
#print(start_sep_pos)
# fill-in missing pos
'''for i in range(1,len(start_sep_pos)):
if start_sep_pos[i] - start_sep_pos[i-1] > scale + 4:
mid = (start_sep_pos[i] + start_sep_pos[i-1]) // 2
good[0:h, mid:mid + 5] = 0
'''
# fill seps start from begin sep with scale space
if len(start_sep_pos) > 0 and len(solid_sep_pos) > 0:
pos_x = start_sep_pos[0]
scale = int(round(1.0 * w / num_cells) + 0.1)
while pos_x < w:
if any(x in solid_sep_pos for x in range(pos_x-3,pos_x+4)):
pos_x = min([x for x in range(pos_x-3,pos_x+4) if x in solid_sep_pos])
good[0:h,pos_x:pos_x+5] = 0
pos_x += scale
else:
# handle special case for 2 cells
if w / scale > 1.5 and w / scale < 2.6:
mid = w // 2
good[0:h, mid:mid + 5] = 0
else:
# fill seps start from solid sep with scale space
proj_x = np.sum(good, axis=0)
mask_x = proj_x > h * 0.9
sep_pos = np.array([i[0] for i in np.argwhere(mask_x)])
pos_x = scale if len(sep_pos) == 0 else sep_pos[0]
while pos_x < w:
good[0:h, pos_x:pos_x + 5] = 0
pos_x += scale + 1
return np.array((1-good) * 255).astype('uint8')
