def model(img, model='keras', adjust=False, detectAngle=False):
"""
@@param:img,
@@param:model,选择的ocr模型，支持keras\pytorch版本
@@param:adjust 调整文字识别结果
@@param:detectAngle,是否检测文字朝向
"""
angle = 0
if detectAngle:
# 进行文字旋转方向检测，分为[0, 90, 180, 270]四种情况
angle = angle_detect(img=np.copy(img)) # 文字朝向检测
print('The angel of this character is:', angle)
im = Image.fromarray(img)
print('Rotate the array of this img!')
if angle == 90:
im = im.transpose(Image.ROTATE_90)
elif angle == 180:
im = im.transpose(Image.ROTATE_180)
elif angle == 270:
im = im.transpose(Image.ROTATE_270)
img = np.array(im)
# 进行图像中的文字区域的识别
text_recs, tmp, img = text_detect(img)
# 识别区域排列
text_recs = sort_box(text_recs)
#
result = crnnRec(img, text_recs, model, adjust=adjust)
return result, tmp, angle
def model(img,adjust=False,detectAngle=False):
"""
@@param:img,
@@param:model,选择的ocr模型，支持keras\pytorch版本
@@param:adjust 调整文字识别结果
@@param:detectAngle,是否检测文字朝向
"""
angle = 0
if detectAngle:
angle = angle_detect(img=np.copy(img))##文字朝向检测
im = Image.fromarray(img)
if angle==90:
im = im.transpose(Image.ROTATE_90)
elif angle==180:
im = im.transpose(Image.ROTATE_180)
elif angle==270:
im = im.transpose(Image.ROTATE_270)
img = np.array(im)
text_recs,tmp,img = text_detect(img)
text_recs = sort_box(text_recs)
result = crnnRec(img,text_recs,model,adjust=adjust)
return result,tmp,angle
def model(img, _model='keras', adjust=False, detect_angle=False):
"""
:param img:
:param _model: 选择的ocr模型，支持keras\pytorch版本
:param adjust: 调整文字识别结果
:param detect_angle: 是否检测文字朝向
:return:
"""
angle = 0
if detect_angle:
# 文字朝向检测
angle = angle_detect(img=np.copy(img))
im = Image.fromarray(img)
if angle == 90:
im = im.transpose(Image.ROTATE_90)
elif angle == 180:
im = im.transpose(Image.ROTATE_180)
elif angle == 270:
im = im.transpose(Image.ROTATE_270)
img = np.array(im)
text_recs, tmp, img = text_detect(img)
text_recs = sort_box(text_recs)
result = crnnRec(img, text_recs, _model, adjust=adjust)
return result, tmp, angle
def dete2img(img, adjust=False):
"""
识别结果单独显示
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('./ctpn/ctpn/text.yml')
text_recs, img_framed, img_scale = text_detect(img)
text_recs = sort_box(text_recs)
result = charRec(img_scale, text_recs, adjust)
h, w = img.shape[:2]
img_table = np.ones([h, w, 3], dtype=img.dtype) * 255
f = getScale(img, scale=TextLineCfg.SCALE, max_scale=TextLineCfg.MAX_SCALE)
for key in result:
for i in range(len(result[key][0])):
result[key][0][i] = int(result[key][0][i] * 1.0 / f)
for key in result:
result[key][1] = charClassification.delSpaceOfChn(result[key][1])
result_copy = sorted(result.items(),
key=lambda x: x[1][0][1],
reverse=False)
chnList = []
for ind in range(len(result_copy)):
chnList.append(result_copy[ind][1][1])
img_table = showChn4table_pil(img_table, chnList)
return result, img_framed, img_table
def modelPicExpansion_v(img, adjust=False):
"""
识别结果在图片下侧显示
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('./ctpn/ctpn/text.yml')
text_recs, img_framed, img_scale = text_detect(img)
text_recs = sort_box(text_recs)
result = charRec(img_scale, text_recs, adjust)
h, w = img.shape[:2]
img_expan = np.zeros([h * 2, w, 3], dtype=img.dtype)
img_expan[0:h, 0:w, :] = img
f = getScale(img, scale=TextLineCfg.SCALE, max_scale=TextLineCfg.MAX_SCALE)
for i, list in enumerate(text_recs):
for j, val in enumerate(list):
text_recs[i][j] = int(val * 1.0 / f)
for key in result:
x1 = result[key][0][0]
y1 = h + result[key][0][1]
x2 = result[key][0][6]
y2 = h + result[key][0][7]
cv2.rectangle(img_expan, (x1, y1), (x2, y2), (255, 255, 255),
thickness=-1)
cv2.rectangle(img_expan, (x1, y1 - h), (x2, y2 - h), (255, 0, 0),
thickness=2)
pilimg = Image.fromarray(img_expan)
draw = ImageDraw.Draw(pilimg)
fs = getFontSize((y2 - y1))
font = ImageFont.truetype("simhei.ttf", fs, encoding="utf-8")
draw.text((x1 + 2, y1 + 4), result[key][1], (0, 0, 0), font=font)
img_expan = np.array(pilimg)
return result, img_expan
def model(img, adjust=False):
cfg_from_file('./ctpn/ctpn/text.yml')
text_recs, img_framed, img = text_detect(img)
text_recs = sort_box(text_recs)
t = time.time()
result = charRec(img, text_recs, adjust)
print('OCR',(time.time()-t))
return result, img_framed
def model(self, img, adjust=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
text_recs, img_framed, img = text_detect(img)
text_recs = sort_box(text_recs)
result = charRec(img, text_recs, adjust)
return result, img_framed
def model(img, adjust=True):
"""
@img: 图片
"""
text_recs, img_framed, img = text_detect(img)
text_recs = sort_box(text_recs)
result = charRec(img, text_recs, adjust)
return result, img_framed
def model(img, adjust=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('./ctpn/ctpn/text.yml')
text_recs, img_framed, img = text_detect(img)
text_recs = sort_box(text_recs)
result = charRec(img, text_recs, adjust)
return result, img_framed
def model(img, adjust=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('./ctpn/ctpn/text.yml')
text_recs, img_framed, img = text_detect(img)
text_recs = sort_box(text_recs)
result = charRec(img, text_recs, adjust)
return result, img_framed
def model(img, model='keras', adjust=False):
"""
@@param:img,
@@param:model,选择的ocr模型，支持keras\pytorch版本
@@param:adjust 调整文字识别结果
@@param:detectAngle,是否检测文字朝向
"""
text_recs, tmp, img = text_detect(img)
text_recs = sort_box(text_recs)
result = crnnRec(img, text_recs, model, adjust=adjust)
return result, tmp
def get_text_region(img, adjust=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('./ctpn/ctpn/text.yml')
text_recs, img_framed, img, detect_time, detect_number = text_detect(img)
#tf.get_default_graph().finalize()
text_recs = sort_box(text_recs)
total_image, total_width = adjust_text_region(img, text_recs, adjust)
return total_image, total_width, img_framed, detect_time, detect_number
def model(img, model='keras', adjust=False, detectAngle=False):
"""
@@param:img,
@@param:model,选择的ocr模型，支持keras\pytorch版本
@@param:adjust 调整文字识别结果
@@param:detectAngle,是否检测文字朝向
"""
# 进行图像中的文字区域的识别
text_recs, tmp, img = text_detect(img)
# 识别区域排列
text_recs = sort_box(text_recs)
#
print(text_recs)
def bolian_model(img, adjust=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('./ctpn/ctpn/text.yml')
# text_recs数据类型是：[[0,1,2,3,4,5,6,7]] (0,1):左上 (2,3):右上 (4,5):左下 (6,7):右下
text_recs, img_framed, img = text_detect(img)
text_recs = sort_box(text_recs)
result = charRec(img, text_recs, adjust)
retData = []
for key in result:
retData.append(result[key][1])
return retData
def model(img, adjust=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('./ctpn/ctpn/text.yml')
sttime = time.time()
text_recs, img_framed, img = text_detect(img)
print("text_detect时间", time.time() - sttime)
# print("test_rec", text_recs)
# print("img_framed", img_framed)
text_recs = sort_box(text_recs)
result = charRec(img, text_recs, adjust)
return result, img_framed
def model(img, adjust=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('./ctpn/ctpn/text.yml')
t = time.time()
text_recs, img_framed, img = text_detect(img)
print("text_detect cost {:.3f}s".format(time.time() - t))
t2 = time.time()
text_recs = sort_box(text_recs)
result = charRec(img, text_recs, adjust)
print("charRec cost {:.3f}s".format(time.time() - t2))
return result, img_framed
def modeltest(img, model='keras', adjust=False, detectAngle=False):
"""
@@param:img,
@@param:model,选择的ocr模型，支持keras\pytorch版本
@@param:adjust 调整文字识别结果
@@param:detectAngle,是否检测文字朝向
"""
angle = 0
# 进行图像中的文字区域的识别
text_recs, tmp, img=text_detect(img)
# 识别区域排列
text_recs = model.sort_box(text_recs)
#
result = model.crnnRec(img, text_recs, model, adjust=adjust)
return result, tmp, angle
def model(img, adjust=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('../ctpn/ctpn/text.yml')
text_recs, img_framed, img = text_detect(img) #检测文本框
# text_recs：文本框位置（列表形式），
# img_framed：在原图上标注了文本框的图片，
# img:原图
text_recs = sort_box(text_recs) #排序框
result = charRec(img, text_recs, adjust) #OCR模型，进行字符识别
return result, img_framed
def model_1(img, adjust=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('./ctpn/ctpn/text.yml')
text_recs, img_framed, img_expan = text_detect(img)
text_recs = sort_box(text_recs)
result = charRec(img_expan, text_recs, adjust)
f = getScale(img, scale=TextLineCfg.SCALE, max_scale=TextLineCfg.MAX_SCALE)
print(f)
for key in result:
for i in range(len(result[key][0])):
result[key][0][i] = int(result[key][0][i] * 1.0 / f)
return result, img_framed
def web_model(img, adjust=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('./ctpn/ctpn/text.yml')
# text_recs数据类型是：[[0,1,2,3,4,5,6,7]] (0,1):左上 (2,3):右上 (4,5):左下 (6,7):右下
text_recs, img_framed, img = text_detect(img)
text_recs = sort_box(text_recs)
result = charRec(img, text_recs, adjust)
retData = []
i = 0
for key in result:
retData.append({
'loc': text_recs[0].tolist(),
'chinese': result[key][1]
})
i = i + 1
return retData
def model(img, adjust=False, detectAngle=False):
angle = 0
if detectAngle:
angle = angle_detect(img=np.copy(img))
im = Image.fromarray(img)
if angle == 90:
im = im.transpose(Image.ROTATE_90)
elif angle == 180:
im = im.transpose(Image.ROTATE_180)
elif angle == 270:
im = im.transpose(Image.ROTATE_270)
img = np.array(im)
text_recs, tmp, img = text_detect(img)
text_recs = sort_box(text_recs)
result = crnnRec(img, text_recs, model, adjust=adjust)
return result, tmp, angle
def model(img, adjust=False, detectAngle=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
angle = 0
if detectAngle:
angle = angle_detect(img=np.copy(img)) ##文字朝向检测
im = Image.fromarray(img)
if angle == 90:
im = im.transpose(Image.ROTATE_90)
elif angle == 180:
im = im.transpose(Image.ROTATE_180)
elif angle == 270:
im = im.transpose(Image.ROTATE_270)
img = np.array(im)
cfg_from_file('./ctpn/ctpn/text.yml')
text_recs, img_framed, img = text_detect(img)
text_recs = sort_box(text_recs)
result = charRec(img, text_recs, adjust)
return result, img_framed
def model(img, adjust=False):
"""
@img: 图片
@adjust: 是否调整文字识别结果
"""
cfg_from_file('./ctpn/ctpn/text.yml')
loops = 10
elapse0 = elapse1 = elapse2 = 0
for v in range(loops):
t0 = time.time()
text_recs, img_framed, img = text_detect(img)
t1 = time.time()
elapse0 += t1 - t0
text_recs = sort_box(text_recs)
t2 = time.time()
elapse1 += t2 - t1
result = charRec(img, text_recs, adjust)
t3 = time.time()
elapse2 += t3 - t2
print("Each: ", (elapse2 / (v + 1)))
return result, img_framed
def model(img, model='keras', adjust=False, detectAngle=False):
angle = 0
if detectAngle:
angle = angle_detect(img=np.copy(img))
print('The angel of this character is:', angle)
im = Image.fromarray(img)
print('Rotate the array of this img!')
if angle == 90:
im = im.transpose(Image.ROTATE_90)
elif angle == 180:
im = im.transpose(Image.ROTATE_180)
elif angle == 270:
im = im.transpose(Image.ROTATE_270)
img = np.array(im)
text_recs, tmp, img = text_detect(img)
text_recs = sort_box(text_recs)
#
result = crnnRec(img, text_recs, model, adjust=adjust)
return result, tmp, angle
def model(img, model, adjust=False, detectAngle=False):
"""
@@param:img,输入的图像数组
@@param:model,选择的ocr模型，支持pytorch版本
@@param:adjust 调整文字识别结果
@@param:detectAngle,是否检测文字朝向
"""
angle = 0
if detectAngle:
# 进行文字旋转方向检测，分为[0, 90, 180, 270]四种情况,下面所用的是逆时针旋转，因此需要改正
angle = angle_detect(img=np.copy(img)) ##文字朝向检测
print('The angel of this character is:', angle)
im = Image.fromarray(img)
print('Rotate the array of this img!')
if angle == 90:
im = im.transpose(Image.ROTATE_270)
elif angle == 180:
im = im.transpose(Image.ROTATE_180)
elif angle == 270:
im = im.transpose(Image.ROTATE_90)
img = np.array(im)
print(img)
# 进行图像中的文字区域的识别
t = time.time()
text_recs, tmp, img = text_detect(img)
print('image area recognition finished!')
print("It takes time:{}s".format(time.time() - t))
# 识别区域排列
text_recs = sort_box(text_recs)
# 文本识别
t = time.time()
result = crnnRec(img, text_recs, ocrMode=model, adjust=adjust)
print('end-to-end text recognition finished!')
print("It takes time:{}s".format(time.time() - t))
return result, tmp, angle
def model(img, model_type='keras_crnn', adjust=False, angle_detect=False):
"""
@img: 图片
@model_type: ocr模型类型
@adjust: 是否调整文字识别结果
@angle_detect: 是否检测文字方向
"""
angle = 0
if angle_detect:
angle = angle_detector(img=np.copy(img)) # 文字方向检测
img = Image.fromarray(img)
if angle == 90:
img = img.transpose(Image.ROTATE_90)
elif angle == 180:
img = img.transpose(Image.ROTATE_180)
elif angle == 270:
img = img.transpose(Image.ROTATE_270)
img = np.array(img)
cfg_from_file('./ctpn/ctpn/text.yml')
text_recs, tmp, img = text_detect(img)
text_recs = sort_box(text_recs)
result = charRec(img, text_recs, model_type, adjust)
return result, tmp, angle
from ctpn.text_detect import text_detect
import ocr
# define test images dir
image_files = glob('./test_images/02*.*')
result_dir = './test_result'
if os.path.exists(result_dir):
shutil.rmtree(result_dir)
os.mkdir(result_dir)
# loop over all images
for image_file in sorted(image_files):
img = np.array(Image.open(image_file).convert('RGB'))
t = time.time()
# detect text area
text_recs, img_framed, img = text_detect(img)
# arrange box from top to bottom, from left to right
text_recs = ocr.sort_box(text_recs)
# load OCR model and recognize
result = ocr.charRec(img, text_recs, False)
# output result
output_file = os.path.join(result_dir, image_file.split('/')[-1])
Image.fromarray(img_framed).save(output_file)
print("Mission complete, it took {:.3f}s".format(time.time() - t))
print("\nRecognition Result:\n")
for key in result:
print(result[key][1])
def model(index, im_name, img, model='keras', adjust=True, recheck_flag=True):
"""
@@param:img,
@@param:model,选择的ocr模型，支持keras\pytorch版本
@@param:adjust 调整文字识别结果
"""
original_img = img
# 进行图像中的文字区域的识别
text_recs, scale, img, abnormal_areas = text_detect(
os.path.join(RESULTS_CTPN,
im_name.split('.')[0] + str(index) + '.png'),
img,
recheck_flag=recheck_flag)
if len(abnormal_areas):
print abnormal_areas[0]
#有异常区域，需要重新检测,暂时只检测第一个
#将异常区域贴到新的图片上，防止异常区域过小
abnormal_img = Image.fromarray(
img[int(abnormal_areas[0][1]):int(abnormal_areas[0][3]),
int(abnormal_areas[0][0]):int(abnormal_areas[0][2]), :])
sub_size = max(abnormal_img.size[0], abnormal_img.size[1])
if sub_size < 200:
sub_img = Image.new('RGB', (200, 200), (255, 255, 255))
else:
sub_img = Image.new('RGB', (sub_size, sub_size), (255, 255, 255))
sub_img.paste(abnormal_img, (0, 0))
sub_img.save('subimage.jpg')
sub_text_recs, sub_scale, sub_img, _ = text_detect(
os.path.join(RESULTS_CTPN,
im_name.split('.')[0] + str(index) + 'sub.png'),
np.array(sub_img.convert('RGB')),
recheck_flag=False)
finally_text_recs = []
# 把大图片和小图片的text_rec合并
# 大图片
for rec in text_recs:
for i in range(4):
if not rec_in_area(rec, abnormal_areas[0], scale):
finally_text_recs.append(rec)
# 小图片
for rec in sub_text_recs:
#绝对位置
abs_rec = []
for i in range(4):
abs_rec.append(
int(rec[2 * i] / scale + abnormal_areas[0][0] / scale))
abs_rec.append(
int(rec[2 * i + 1] / scale + abnormal_areas[0][1] / scale))
finally_text_recs.append(abs_rec)
else:
finally_text_recs = text_recs
draw_box(im_name, original_img, finally_text_recs)
# 识别区域排列
# text_recs = sort_box(text_recs)
#
# results = crnnRec(im_name, original_img, text_recs, model, adjust=adjust)
results = crnnRec(im_name,
original_img,
finally_text_recs,
model,
adjust=adjust)
# draw new image
new_img = np.ones(original_img.shape) * 255
# new_img = np.ones(IMAGE_SIZE) * 255
color_ = (0, 0, 0)
text_size = 20
ft = draw.put_chinese_text(
'/datasets/text_renderer/data/fonts/chn/songti.ttf')
f_result = open('results/{}.txt'.format(im_name), 'w')
f_json = open('results/{}.json'.format(im_name.split('.')[0]), 'w')
result_dict = []
for result in results.items():
box, b = result[1]
pos = (box[0], box[1])
print b[0]
#print box
new_img = ft.draw_text(new_img, pos, b[0], text_size, color_)
#f_result.write('{} {}, {} \n'.format(b[0], pos[0], pos[1]))
result_dict.append(
[box[0], box[1], box[2] - box[0], box[5] - box[3], b[0]])
cv2.imwrite(os.path.join(RESULTS, im_name), new_img)
f_json.write(json.dumps(result_dict, ensure_ascii=False))
return
def model(img,model='keras',adjust=False):
text_recs,tmp,img = text_detect(img)
text_recs = sort_box(text_recs)
result = crnnRec(img,text_recs,model,adjust=adjust)
return result,tmp
0] > partImg.shape[1]: # 过滤异常图片
continue
image = Image.fromarray(partImg).convert('L')
text = do_predict(rec_model, image)
if len(text) > 0:
results[index] = [rec]
results[index].append(text) # 识别文字
return results
if __name__ == '__main__':
result_dir = './test_result'
if os.path.exists(result_dir):
shutil.rmtree(result_dir)
os.mkdir(result_dir)
# rec_model =
rec_model = keras.models.load_model('/workzone/tmp/basemodel.h5')
for image_file in sorted(image_files):
image = np.array(Image.open(image_file).convert('RGB'))
t = time.time()
text_recs, img_framed, image = text_detect(image)
text_recs = sort_box(text_recs)
result = charRec(rec_model, image, text_recs, adjust=False)
pass
