def crnnRec(im, boxes, leftAdjust=False, rightAdjust=False, alph=0.2, f=1.0):
"""
crnn模型，ocr识别
@@model,
@@converter,
@@im:Array
@@text_recs:text box
@@ifIm:是否输出box对应的img
"""
results = []
im = Image.fromarray(im)
for index, box in enumerate(boxes):
degree, w, h, cx, cy = solve(box)
partImg, newW, newH = rotate_cut_img(im, degree, box, w, h, leftAdjust,
rightAdjust, alph)
newBox = xy_rotate_box(cx, cy, newW, newH, degree)
# partImg_ = partImg.convert('L')
# simPred = crnnOcr(partImg_)##识别的文本
# if simPred.strip()!=u'':
# results.append({'cx':cx*f,'cy':cy*f,'text':simPred,'w':newW*f,'h':newH*f,'degree':degree*180.0/np.pi})
results.append({
'cx': cx * f,
'cy': cy * f,
'text': '',
'w': newW * f,
'h': newH * f,
'degree': degree * 180.0 / np.pi
})
return results
def ocr_batch(self, img, boxes, leftAdjustAlph=0.0, rightAdjustAlph=0.0):
"""
batch for ocr
"""
im = Image.fromarray(img)
newBoxes = []
for index, box in enumerate(boxes):
partImg, box = rotate_cut_img(im, box, leftAdjustAlph,
rightAdjustAlph)
box['img'] = partImg.convert('L')
newBoxes.append(box)
res = self.ocrModel(newBoxes)
return res
def crnnRec(im, boxes, leftAdjust=False, rightAdjust=False, alph=0.2, f=1.0):
"""
crnn模型，ocr识别
leftAdjust,rightAdjust 是否左右调整box 边界误差，解决文字漏检
"""
results = []
im = Image.fromarray(im)
for index, box in enumerate(boxes):
degree, w, h, cx, cy = solve(box)
partImg, newW, newH = rotate_cut_img(im, degree, box, w, h, leftAdjust, rightAdjust, alph)
text = crnnOcr(partImg.convert('L'))
if text.strip() != u'':
results.append({'cx': cx * f, 'cy': cy * f, 'text': text, 'w': newW * f, 'h': newH * f,
'degree': degree * 180.0 / np.pi})
return results
def crnnRec(im,
boxes,
leftAdjust=False,
rightAdjust=False,
alph=0.2,
f=1.0,
save=True):
"""
crnn模型，ocr识别
leftAdjust, rightAdjust 是否左右调整box 边界误差，解决文字漏检
"""
results = []
# im = Image.fromarray(im)
im = Image.fromarray(cv2.cvtColor(im, cv2.COLOR_BGR2RGB))
for index, box in enumerate(boxes):
degree, w, h, cx, cy = solve(box)
# 按照box大小，裁剪图片
partImg, newW, newH = rotate_cut_img(im, degree, box, w, h, leftAdjust,
rightAdjust, alph)
# partImg, newW, newH = cut_img = ()
image_cv = cv2.cvtColor(numpy.asarray(partImg), cv2.COLOR_RGB2BGR)
cv2.imshow("crnnRec", image_cv)
# partImg = Image.fromarray(cv2.cvtColor(partImg, cv2.COLOR_BGR2RGB))
# 图片会转灰度图，进行识别
# print('crnnRec', partImg.size)
text, w = crnnOcr(partImg.convert('L'))
# text, w = crnnOcr(partImg)
if save:
image = partImg.resize((w, 32), Image.BILINEAR)
save_image(image, text)
# text = crnnOcr(partImg)
if text.strip() != u'':
results.append({
'cx': cx * f,
'cy': cy * f,
'text': text,
'w': newW * f,
'h': newH * f,
'degree': degree * 180.0 / np.pi
})
return results
def ocr_batch(self, img, boxes, leftAdjustAlph=0.0, rightAdjustAlph=0.0):
"""
batch for ocr
"""
im = Image.fromarray(img)
newBoxes = []
for index, box in enumerate(boxes):
partImg, box = rotate_cut_img(im, box, leftAdjustAlph,
rightAdjustAlph) # 旋转裁切图片。
box['img'] = partImg.convert(
'L'
) # L = R * 299/1000 + G * 587/1000+ B * 114/1000 转换。看下面这个，我明明转换成L了，为什么输出的图像还有颜色呢？不是应该是灰度图吗？
newBoxes.append(box)
res = self.ocrModel(newBoxes)
return res
def ocr_batch(self, img, boxes, leftAdjustAlph=0.0, rightAdjustAlph=0.0):
"""
batch for ocr
"""
im = Image.fromarray(img)
newBoxes = []
for index, box in enumerate(boxes):
partImg, box = rotate_cut_img(im, box, leftAdjustAlph,
rightAdjustAlph)
img = np.array(partImg)
_, img_bright = cv.threshold(img, 200, 255, cv.THRESH_BINARY)
# cv.imshow('mg0',img_bright)
# cv.waitKey()
box['img'] = partImg.convert('L')
newBoxes.append(box)
res = self.ocrModel(newBoxes)
return res
def crnnRec(im,
boxes,
leftAdjust=False,
rightAdjust=False,
alph=0.2,
f=1.0,
tp_groups=None,
boxAll=None,
scoreAll=None):
"""
crnn模型，ocr识别
leftAdjust,rightAdjust 是否左右调整box 边界误差，解决文字漏检
"""
results = []
im = Image.fromarray(im)
for index, box in enumerate(boxes):
degree, w, h, cx, cy = solve(box)
partImg, newW, newH = rotate_cut_img(im, degree, box, w, h, leftAdjust,
rightAdjust, alph)
text = crnnOcr(partImg.convert('L'))
detailbox = boxAll[tp_groups[index]]
detailscore = scoreAll[tp_groups[index]]
detaildex = tp_groups[index]
if text.strip() != u'':
results.append({
'cx': cx * f,
'cy': cy * f,
'text': text,
'w': newW * f,
'h': newH * f,
'degree': degree * 180.0 / np.pi,
'detailbox': detailbox,
'detailscore': detailscore,
'detaildex': detaildex
})
#degree表示顺时针转多少度.
return results
config['img'] = img
text_recs = text_detect(**config)
# print(text_recs)
boxes = sorted(text_recs, key=lambda x: sum([x[1], x[3], x[5], x[7]]))
i = 0
filename = ntpath.basename(imgPath)
ori_filename = romvChinese(filename)
for index, box in enumerate(boxes):
filename = ori_filename
degree, w, h, cx, cy = solve(box)
partImg, newW, newH = rotate_cut_img(img,
degree,
box,
w,
h,
leftAdjust=True,
rightAdjust=True,
alph=0.2)
if partImg.size[1] < 32:
scale = partImg.size[1] * 1.0 / 32
w = partImg.size[0] / scale
w = int(w)
partImg = partImg.resize((w, 32), Image.BILINEAR)
filename = filename[:-4] + '_' + str(i) + '_.jpg'
partImgPath = os.path.join(subdir, filename)
partImg.save(partImgPath)
i += 1
img_count += 1
# print('img_count: ', img_count)
if img_count % subdir_interval == 0: