def get_image_box(gray):
bw = imu.preprocess_bw_inv(gray)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (49, 13))
bw2 = cv2.dilate(bw, kernel, iterations=3)
image, contours, hierarchy = cv2.findContours(bw2, cv2.RETR_TREE,
cv2.CHAIN_APPROX_SIMPLE)
boxes = []
# cv2.imwrite('temp.png', bw2)
cv2.namedWindow("bw2", cv2.WINDOW_AUTOSIZE)
cv2.imshow("bw2", bw2)
cv2.waitKey(0)
cv2.destroyAllWindows()
for i, cnt in enumerate(contours):
area = cv2.contourArea(cnt)
rect = cv2.minAreaRect(cnt)
box = cv2.boxPoints(rect)
box = np.int0(box)
height = abs(box[0][1] - box[2][1])
width = abs(box[0][0] - box[2][0])
boxList = box.tolist()
origin = min(boxList)
# print(box)
if (height > 150):
boxes.append(
(origin[0], origin[1], origin[0] + width, origin[1] + height))
# exit(0)
return boxes
def get_vertical_white_edge_dis(img):
"""Get vertical white edge distance of character image.
Args:
img: Character image.
Returns:
An interger distance.
"""
bw = imu.preprocess_bw_inv(img)
weight, profile = imu.horizontal_proj(bw)
horCoorY1 = 0
if weight[0] == 0:
i = 0
while weight[i] == 0 and i < len(weight):
i += 1
horCoorY1 = i
horCoorY2 = len(weight) - 1
if weight[-1] == 0:
j = len(weight) - 1
while weight[j] == 0 and j > 0:
j -= 1
horCoorY2 = j
return horCoorY1, len(weight) - horCoorY2
def text(self, options={}):
imgInv = imu.preprocess_bw_inv(self.image_gray)
cv2.imwrite('imgInv.jpg', imgInv)
connected_components = get_cc('imgInv.jpg')
os.remove('imgInv.jpg')
connected_components = [
length2coor(cc) for cc in connected_components
if 5 < cc[2] < 60 and 3 < cc[3] < 60
]
# word_boxes = generate_words(connected_components, 50)
connected_components = generate_chars(connected_components)
word_boxes = generate_lines(connected_components, 50)
for word_box in word_boxes:
word_result = self.text_word(word_box)
keys = list(word_result.keys())
keys.sort(key=lambda item: item[0])
for key in keys:
result = word_result[key]
if options.get("probability") == "true":
print(f'({result["char"]}, {result["confidence"]})',
end=' ')
else:
print(f'{result["char"]}', end=' ')
# if result["char"]=="二":
# cv2.imwrite(f"二{now}.png", self.image_gray[key[1]:key[3], key[0]:key[2]])
print()
def divide_ver(img):
"""Divide paper by blank detection in vertical direction.
Args:
img: Image to be divided.
Returns:
A list of horizontal coordinates.
"""
verCoorX1 = []
verCoorX2 = []
temp = []
temp1 = []
temp2 = []
result = []
bw = imu.preprocess_bw_inv(img)
weight, profile = imu.vertical_proj(bw)
# print('weight = {}'.format(weight[:100]))
for i in range(0, len(weight)):
if weight[i] > 0:
if i == len(weight) - 1 or weight[i + 1] == 0:
verCoorX2.append(i)
if i == 0 or weight[i - 1] == 0:
verCoorX1.append(i)
# print(verCoorX1)
# print(verCoorX2)
temp = list(zip(verCoorX1, verCoorX2))
# print(temp)
for r in temp:
if r[1] - r[0] > 5:
temp1.append(r)
first = 1 if temp1[0][0] < 100 else 0
start = temp1[first][0]
end = temp1[first][1]
for r in range(first, len(temp1)):
if r == len(temp1) - 1:
end = temp1[r][1]
temp2.append((start, end))
elif temp1[r + 1][0] - temp1[r][1] < 45:
continue
else:
end = temp1[r][1]
temp2.append((start, end))
start = temp1[r + 1][0]
for i in temp2:
if i[1] - i[0] > 800:
result.append(i)
if len(result) > 2:
result = result[0:2]
# print(result)
# print(temp)
return result
def find_char(img, coor):
"""Divide line by blank detection in horizontal direction from left to right.
Args:
img: Image to be divided.
coor: A list of line coordinates.
Returns:
A list of first 2 or 3 characters coordinates of each lines.
"""
result = []
for i, c in enumerate(coor):
verCoorX1 = []
verCoorX2 = []
line = []
imgLine = img[c[1]:c[3], c[0]:c[2]]
bw = imu.preprocess_bw_inv(imgLine, smooth=False)
# cv2.imwrite('./res/lines/'+str(i)+'.png', bw)
weight, profile = imu.vertical_proj(bw)
if weight[0] > 1:
verCoorX1.append(0)
for i in range(1, len(weight) - 1):
if weight[i] > 1:
if weight[i + 1] <= 1:
verCoorX2.append(i)
if weight[i - 1] <= 1:
verCoorX1.append(i)
if weight[-1] > 1:
verCoorX2.append(len(weight) - 1)
# print(verCoorX1)
# print(verCoorX2)
ver = list(zip(verCoorX1, verCoorX2))
n = 3
for v in ver:
if n == 0:
break
elif v[1] - v[0] < 3:
continue
else:
line.append((v[0] + c[0], c[1], v[1] + c[0], c[3]))
n -= 1
result.append(line)
return result
def divide_hor(img, coor):
"""Divide paper by blank detection in horizontal direction.
Args:
img: Image to be divided.
Returns:
A list of vertical coordinates.
"""
h, w = img.shape[0:2]
result = []
for c in coor:
horCoorY1 = []
horCoorY2 = []
imgPart = img[0:h, c[0]:c[1]]
bw = imu.preprocess_bw_inv(imgPart)
weight, profile = imu.horizontal_proj(bw)
if weight[0] > 0:
horCoorY1.append(0)
for i in range(1, len(weight) - 1):
if weight[i] > 0:
if weight[i + 1] == 0:
horCoorY2.append(i)
if weight[i - 1] == 0:
horCoorY1.append(i)
if weight[-1] > 0:
horCoorY2.append(len(weight) - 1)
# print(verCoorX1)
# print(verCoorX2)
line = list(zip(horCoorY1, horCoorY2))
for l in line:
if l[1] - l[0] > 10:
result.append((c[0], l[0], c[1], l[1]))
return result
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 get_dilation(gray):
bw = imu.preprocess_bw_inv(gray)
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (49, 13))
bw2 = cv2.dilate(bw, kernel, iterations=1)
return bw2
def deal_one_page(srcPath,
desPath='',
stanPath='',
saveImage=True,
charsOnly=False,
rectification=False):
"""
"""
global heightOfLine
coor5 = []
mistakes = 0
bdResult = {}
name = os.path.basename(srcPath)
if not os.path.exists(srcPath):
print("Image path not exists!")
return None
try:
imgBgr = cv2.imread(srcPath)
if rectification:
imgBgr = rectify(imgBgr.copy())
imgData = cv2.cvtColor(imgBgr, cv2.COLOR_BGR2GRAY)
# imgData = cv2.imread(srcPath, cv2.IMREAD_GRAYSCALE)
# imgBgr = cv2.cvtColor(imgData, cv2.COLOR_GRAY2BGR)
except Exception as imageError:
print(imageError, 'Could not read the image file, skip...')
# cv2.namedWindow("imgData", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("imgData", imgData)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# exit(0)
imgEliVer = eli_ver(imgData.copy()) #
# cv2.namedWindow("imgEliVer", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("imgEliVer", imgEliVer)
coor1 = divide_ver(imgEliVer) #
# print('coor1: {}'.format(coor1))
if heightOfLine == 0: #
coor2 = divide_hor(imgEliVer, coor1)
heightOfLines = [c[3] - c[1] for c in coor2]
heightOfLine = median(heightOfLines)
imgInv = imu.preprocess_bw_inv(imgData.copy())
cv2.imwrite('imgInv.jpg', imgInv)
coorCC = get_no_intersect_boxes('imgInv.jpg')
imgEliCC = eli_large_cc(coorCC, imgData.copy(), heightOfLine)
os.remove('imgInv.jpg')
# cv2.namedWindow("imgEliCC", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("imgEliCC", imgEliCC)
imgEli = eli_ver(imgEliCC) #1
# cv2.namedWindow("imgEli", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("imgEli", imgEli)
coor2 = divide_hor(imgEli, coor1) #2
# print('coor2: {}'.format(coor2))
# img2 = mark_box(imgBgr, coor2, color=(0,255,0))
# cv2.namedWindow("img2", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("img2", img2)
coor3 = find_char(imgEli, coor2) #3
# print(np.array(coor3))
# for l in coor3:
# img3 = mark_box(imgBgr, l, color=(0,255,0))
# cv2.namedWindow("img3", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("img3", img3)
if charsOnly:
return coor3
# print('SVM classifying...')
coor4 = svm_classify(imgEli, coor3) #4
# print(*coor4, sep='\n')
imgBw = imu.preprocess_bw(imgEli, boxSize=(4, 4), morph=False) #re
coor3Bw = find_char(imgBw, coor2)
coor4Bw = svm_classify(imgBw, coor3Bw)
coor5 = update_result(coor4, coor4Bw)
try:
# imgBd = preprocess_bw(imgEliCC)
# print('BaiDu recognizing...')
bdResult = bd_rec(imgEliCC,
client,
imgForm=srcPath[-4:],
api='general')
# except ConnectionResetError as e:
# print('Connection reset by peer, repeat BaiDu recognizion...')
# bdResult = bd_rec(imgEliCC, client, api='accurate')
except Exception as e:
print(
'Baidu recognition error, check your internet connection. exit...')
exit(0)
# print(bdResult)
addResult = additional_questions(bdResult)
# print(addResult)
if addResult:
coor5.append(([addResult], 2))
# print(*coor4, sep='\n')
if stanPath:
stanResult = output_stan(coor5, os.path.basename(srcPath))
mistakes = verification(stanResult, stanPath)
# print(mistakes)
# if saveImage:
# for l in coor5: #5
# # print(l)
# color = (255,255,0)
# if l[1] != 0:
# color = (255,0,0) if l[1] == 2 else (0,255,0)
# img5 = mark_box(imgBgr, l[0], color)
# cv2.imwrite(desPath, img5) #6
# show_image(img5)
# exit(0)
# cv2.namedWindow("img5", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("img5", img5)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# imgEliVer2 = detect_vertical_line2(imgData)
# coor2 = divide_hor(imgEliVer2, coor1)
sa = detect_self_assessment(bdResult)
if sa:
# imgEliSa = imgData.copy()
imgData[sa:, coor1[1][0]:coor1[1][1] + 10] = 255
# coor22 = divide_hor(imgEliSa, coor1)
coor22 = divide_hor(imgData, coor1)
region = integrate_lines(coor5, coor22, imgData.shape[1], imgData.shape[0])
# print('region: {}'.format(region))
# imgEliVerBgr = cv2.cvtColor(imgEliVer2, cv2.COLOR_GRAY2BGR)
# imgRegion = mark_box(imgBgr, region, color=(0,255,0))
# cv2.imwrite(desPath, imgRegion)
# cv2.imwrite('imgInv.jpg', get_dilation(imgData.copy()))
# coorCCDil = get_no_intersect_boxes('imgInv.jpg')
# os.remove('imgInv.jpg')
'''
CCAndType = []
wordBoxes = [w["location"] for w in bdResult["words_result"]]
idxWords = index.Index()
for j,b in enumerate(wordBoxes):
idxWords.insert(j, (b["left"], b["top"], b["left"]+b["width"], b["top"]+b["height"]))
idxRegion = index.Index()
for j,r in enumerate(region):
idxRegion.insert(j, r)
idxCC = index.Index()
for j,c in enumerate(coorCC):
idxCC.insert(j, (c[0], c[1], c[0]+c[2], c[1]+c[3]))
boxOfSpecialWord = get_match_location(bdResult, keyWordsConfig)
hitsOfCC = []
for j,s in enumerate(boxOfSpecialWord):
b = (s[0], s[1], s[0]+s[2], s[1]+s[3])
hitsOfRegion = list(idxRegion.intersection(b, objects=True))
items = [item for item in hitsOfRegion if item.bbox[1] < s[1]+s[3]/2 < item.bbox[3]]
for item in items:
boxOfGraph = (item.bbox[0], b[3]+10, item.bbox[2]-item.bbox[0], item.bbox[3]-b[3]-20)
boxOfGraph1 = (item.bbox[0], b[3]+10, item.bbox[2], item.bbox[3]-10)
coorCC.append(boxOfGraph)
hitsOfCC += list(idxCC.intersection(boxOfGraph1, objects=False))
# print(hitsOfCC)
coorCCCopy = coorCC.copy()
coorCC = [coorCCCopy[i] for i in range(len(coorCCCopy)) if i not in hitsOfCC]
boxRest = []
for i,c in enumerate(coorCC.copy()):
r = (c[0], c[1], c[0]+c[2], c[1]+c[3])
hitsOfRegion = list(idxRegion.intersection(r))
if len(hitsOfRegion) == 0:
continue
if c[2] > heightOfLine*3 or c[3] > heightOfLine*1.5:
CCAndType.append((c, 1))
# del(coorCC[i])
continue
elif c[2] < heightOfLine*0.5 and c[3] < heightOfLine*0.5:
CCAndType.append((c, 0))
continue
hits = list(idxWords.intersection(r, objects=True))
if hits:
flag = False
for item in hits:
wordBox = (wordBoxes[item.id]["left"], wordBoxes[item.id]["top"], wordBoxes[item.id]["width"], wordBoxes[item.id]["height"])
if is_within(c, wordBox):
CCAndType.append((c, 0))
flag = True
# del(coorCC[i])
continue
if flag:
continue
boxRest.append(c)
typesOfCC = tell_me_text_or_graph(imgBgr.copy(), imgData.copy(), boxRest)
CCAndType += zip(boxRest, typesOfCC)
resultFinal = get_items(region, imgData.copy(), bdResult, CCAndType, name)
'''
imgData1 = detect_vertical_line3(imgData.copy())
resultFinal = get_items_v02(region, imgData1.copy(), coorCC, name)
# coorRegion = []
# coorGraph = []
# coorText = []
# resultFinalA = absolute_coor(resultFinal)
# for x in resultFinalA["questions"]:
# coorRegion.append(dic2coor(x["location"]))
# coorGraph.extend([dic2coor(i["location"]) for i in x["content"] if i["type"]=="graph"])
# coorText.extend([dic2coor(i["location"]) for i in x["content"] if i["type"]=="text"])
# imgRegion = mark_box(imgBgr, coorRegion, color=(0,255,0))
# imgGraph = mark_box(imgRegion, coorGraph, color=(0,0,255))
# imgText = mark_box(imgGraph, coorText, color=(255,255,0))
if desPath:
with open(desPath[:-4] + '.json', 'w') as f:
json.dump(resultFinal, f)
# if saveImage:
# cv2.imwrite(desPath, imgText)
# cv2.namedWindow("imgData", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("imgData", imgData)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
#return coor5, mistakes
return resultFinal
def deal_one_page(srcPath, desPath='', charsOnly=False, rectification=False):
"""Process a paper image.
Args:
srcPath: File path of image.
desPath: Destination path.
charsOnly: Whether only return chars coordinate result.
rectification: Whether rectify image.
Returns:
Final result.
"""
global heightOfLine
coor5 = []
mistakes = 0
bdResult = {}
ratio = 0
name = os.path.basename(srcPath)
if not os.path.exists(srcPath):
print("Image path not exists!")
return None
try:
imgBgr = cv2.imread(srcPath)
if rectification:
imgBgr = rectify(imgBgr.copy())
imgData = cv2.cvtColor(imgBgr, cv2.COLOR_BGR2GRAY)
except Exception as imageError:
print(imageError, 'Could not read the image file, skip...')
# 根据图像长宽判断是否为单栏
ori_size = imgData.shape[0:2]
single_column = True if ori_size[0] > ori_size[1] else False
# 缩放到固定大小(双栏宽度为4000pixels, 单栏为2000pixels)
if single_column and ori_size[1] != 2000:
ratio = 2000 / ori_size[1]
elif not single_column and ori_size[1] != 4000:
ratio = 4000 / ori_size[1]
if ratio:
imgData = resize(imgData, x=ratio)
new_size = imgData.shape[0:2]
#原始图像去除竖线（长度大于100pixels，可能去除部分题目信息）
imgEliVer = eli_ver(imgData.copy()) #
# cv2.namedWindow("imgEliVer", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("imgEliVer", imgEliVer)
#纵向划分试题区域
coor1 = divide_ver(imgEliVer)
# print('coor1: {}'.format(coor1))
#计算行高（所有行高的中位数）
if heightOfLine == 0: #
coor2 = divide_hor(imgEliVer, coor1)
heightOfLines = [c[3] - c[1] for c in coor2]
heightOfLine = median(heightOfLines)
#获取原始图像联通区域以及消去较大的联通区域
imgInv = imu.preprocess_bw_inv(imgData.copy())
cv2.imwrite('imgInv.jpg', imgInv)
coorCC = get_no_intersect_boxes('imgInv.jpg')
imgEliCC = eli_large_cc(coorCC, imgData.copy(), heightOfLine)
os.remove('imgInv.jpg')
# cv2.namedWindow("imgEliCC", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("imgEliCC", imgEliCC)
#去除较大联通区域的图像上去除竖线
imgEli = eli_ver(imgEliCC) #1
# cv2.namedWindow("imgEli", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("imgEli", imgEli)
#行划分
coor2 = divide_hor(imgEli, coor1) #2
# print('coor2: {}'.format(coor2))
# img2 = mark_box(imgBgr, coor2, color=(0,255,0))
# cv2.namedWindow("img2", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("img2", img2)
#获取每行前三个字符
coor3 = find_char(imgEli, coor2) #3
# print(np.array(coor3))
# for l in coor3:
# l = [length2coor(list(map(lambda x: x / ratio, coor2length(c)))) for c in l]
# img3 = mark_box(imgBgr, l, color=(0,255,0))
# cv2.namedWindow("img3", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("img3", img3)
# cv2.imwrite("img3.jpg", img3)
if charsOnly:
coor_chars = []
#坐标换算
for l in coor3:
coor_chars.append([
length2coor(list(map(lambda x: int(x / ratio),
coor2length(c)))) for c in l
])
return coor_chars
#SVM分类字符图像
# print('SVM classifying...')
coor4 = svm_classify(imgEli, coor3) #4
# print(*coor4, sep='\n')
#在二值图上重新获取字符以及SVM分类（补全部分漏检）
imgBw = imu.preprocess_bw(imgEli, boxSize=(4, 4), morph=False) #re
coor3Bw = find_char(imgBw, coor2)
coor4Bw = svm_classify(imgBw, coor3Bw)
coor5 = update_result(coor4, coor4Bw)
#百度识别
try:
# print('BaiDu recognizing...')
bdResult = bd_rec(imgEliCC,
client,
imgForm=srcPath[-4:],
api='general')
except Exception as e:
print(
'Baidu recognition error, check your internet connection. exit...')
exit(0)
# print(bdResult)
#附加题处理
addResult = additional_questions(bdResult)
# print(addResult)
if addResult:
coor5.append(([addResult], 2))
# print(*coor4, sep='\n')
#自我评测处理
sa = detect_self_assessment(bdResult)
if sa:
imgData[sa:, coor1[1][0]:coor1[1][1] + 10] = 255
coor22 = divide_hor(imgData, coor1)
# print(coor22)
#生成题目区域
region = integrate_lines(coor5, coor22, new_size)
# print('region: {}'.format(region))
#去除装订线及中间分割线（不会去除题目信息）
imgData1 = detect_vertical_line3(imgData.copy())
#导出最后结果
resultFinal = get_items_v02(region, imgData1.copy(), coorCC, name)
if desPath:
with open(desPath[:-4] + '.json', 'w') as f:
json.dump(resultFinal, f)
# cv2.namedWindow("imgData", cv2.WINDOW_AUTOSIZE)
# cv2.imshow("imgData", imgData)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
return resultFinal
def testpaper(file_path, save_path):
"""Process a paper image.
Args:
file_path: File path of image.
save_path: Destination path.
Returns:
Final result.
"""
name = os.path.basename(file_path)
image = cv2.imread(file_path)
if image is None:
print('ImageError: Could not read the image file, exit...')
return None
image_gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# 根据图像长宽判断是否为单栏
ori_size = image_gray.shape[0:2]
single_column = True if ori_size[0] > ori_size[1] else False
# 缩放到固定大小(双栏宽度为4000pixels, 单栏为2000pixels)
if single_column and ori_size[1] != 2000:
ratio = 2000 / ori_size[1]
elif not single_column and ori_size[1] != 4000:
ratio = 4000 / ori_size[1]
else:
ratio = 0
if ratio:
image_gray_scale = resize(image_gray, x=ratio)
new_size = image_gray_scale.shape[0:2]
img_bw_inv = imu.preprocess_bw_inv(image_gray_scale)
img_bw = imu.preprocess_bw(image_gray_scale)
img_bw2 = imu.preprocess_bw(image_gray_scale, boxSize=(5, 5))
#原始图像去除竖线（长度大于100pixels，可能去除部分题目信息）
imgEliVer = eli_ver(img_bw2)
#纵向划分试题区域
coor1 = divide_ver(imgEliVer)
if not coor1:
print('coor1: {}'.format(coor1))
print("Can't divide vertically, skip...")
return None
#计算行高（所有行高的中位数）
coor2 = divide_hor(imgEliVer, coor1)
if coor2:
heightOfLines = [c[3] - c[1] for c in coor2]
heightOfLine = median(heightOfLines)
else:
print("Can't divide by line, skip...")
return None
#获取原始图像联通区域以及消去较大的联通区域
imgInv = imu.preprocess_bw_inv(img_bw2.copy())
cv2.imwrite('imgInv.jpg', imgInv)
coorCC = get_no_intersect_boxes('imgInv.jpg')
imgEliCC = eli_large_cc(coorCC, img_bw2.copy(), heightOfLine)
os.remove('imgInv.jpg')
#去除较大联通区域的图像上去除竖线
imgEli = eli_ver(imgEliCC)
#行划分
coor2 = divide_hor(imgEli, coor1)
#获取每行前三个字符
coor3 = find_char(img_bw, coor2)
#SVM分类字符图像
coor4 = svm_classify(img_bw, coor3)
#在二值图上重新获取字符以及SVM分类（补全部分漏检）
imgBw = imu.preprocess_bw(imgEli, boxSize=(4, 4), morph=False)
coor3Bw = find_char(imgBw, coor2)
coor4Bw = svm_classify(imgBw, coor3Bw)
coor5 = update_result(coor4, coor4Bw)
# 百度文字识别API客户端
client = bd_access()
try:
bdResult = bd_rec(imgEliCC, client, imgForm=name[-4:], api='general')
except Exception as e:
print(e, 'Baidu recognition error, exit...', sep='\n')
return None
# 附加题处理
addResult = additional_questions(bdResult)
if addResult:
coor5.append(([addResult], 2))
#自我评测处理
sa = detect_self_assessment(bdResult)
if sa:
image_gray_scale[sa:, coor1[1][0]:coor1[1][1] + 10] = 255
coor22 = divide_hor(image_gray_scale, coor1)
# 生成题目区域
region = integrate_lines(coor5, coor22, new_size)
region = [
length2coor(list(map(lambda x: int(x / ratio), coor2length(r))))
for r in region
]
#背景标记
region_close = []
for r in region:
if r[2] <= r[0] or r[3] <= r[1]: continue
border = imu.strip_white_boder2(image_gray[r[1]:r[3], r[0]:r[2]])
r = num2int([
border[0] + r[0], border[1] + r[1], border[2] + r[0],
border[3] + r[1]
])
region_close.append(r)
background = np.zeros((r[3] - r[1], r[2] - r[0], 3), np.uint8)
background[:] = (200, 170, 150)
image[r[1]:r[3],
r[0]:r[2]] = cv2.addWeighted(image[r[1]:r[3], r[0]:r[2]], 0.7,
background, 0.3, 0)
# image_process = np.concatenate([image_gray_scale, img_bw,
# imgEliVer, imgEliCC, imgEli], axis=0)
# cv2.imwrite("image_process.jpg", image_process)
dic_result = {"positions": region_close}
print(f'region: {region_close}')
print('output to directory {}'.format(save_path))
with open(save_path + '/' + name.replace('.jpg', '.json'), 'w') as f:
json.dump(dic_result, f)
cv2.imwrite(os.path.join(save_path, name), image)
return region
#去除装订线及中间分割线（不会去除题目信息）
imgData1 = detect_vertical_line3(imgData.copy())
#导出最后结果
resultFinal = get_items_v02(region, imgData1.copy(), coorCC, name)
if desPath:
with open(desPath[:-4] + '.json', 'w') as f:
json.dump(resultFinal, f)
return resultFinal