개정판 e266373b
fixed issue #478:
- change logic to extract text from image
| DTI_PID/DTI_PID/tesseract_ocr_module.py | ||
|---|---|---|
| 170 | 170 |
return textInfoList |
| 171 | 171 |
|
| 172 | 172 |
|
| 173 |
'''
|
|
| 173 |
"""
|
|
| 174 | 174 |
@history 2018.04.26 Jeongwoo Make TextInfo object with Calculated Coords (with BoundBox Coords) |
| 175 | 175 |
2018.04.30 Jeongwoo Add QRect.setHeight() in if-statement [(lineRect is not None and currentRect is not None) and lineRect.intersects(currentRect)] |
| 176 | 176 |
2018.05.09 Jeongwoo Check split text' length |
| ... | ... | |
| 179 | 179 |
2018.06.08 Jeongwoo Add angle Parameter and get rotated image's coords |
| 180 | 180 |
2018.06.14 Jeongwoo Add try-except. If exception occured, return None |
| 181 | 181 |
2018.06.20 Jeongwoo Remove variable [lastCharHeight] / Change variable [cey], [ch] / Change method to calculate text line height |
| 182 |
''' |
|
| 182 |
humkyung 2018.10.12 change logic to extract text which first get bounding box and then extract character |
|
| 183 |
""" |
|
| 183 | 184 |
def getTextInfo(img, startPoint, angle = 0, flag = FLAG_IMAGE_TO_BOXES, conf = DEFAULT_CONF): |
| 184 | 185 |
try: |
| 186 |
textInfoList = [] |
|
| 187 |
|
|
| 185 | 188 |
docData = AppDocData.instance() |
| 186 | 189 |
configs = docData.getConfigs('Text Size', 'Min Text Size')
|
| 187 | 190 |
minSize = int(configs[0].value) if 1 == len(configs) else 30 |
| ... | ... | |
| 192 | 195 |
im = im.rotate(-angle, expand=True) |
| 193 | 196 |
imgWidth = im.width |
| 194 | 197 |
imgHeight = im.height |
| 195 |
textOnlyOcrData = pytesseract.image_to_string(im, config=conf, lang='eng') |
|
| 196 |
textInfoList = [] |
|
| 197 |
ADJUST = 2 |
|
| 198 |
if textOnlyOcrData: |
|
| 199 |
boundaryOcrData = pytesseract.image_to_boxes(im, config=conf, lang='eng') |
|
| 200 |
text = textOnlyOcrData |
|
| 201 |
text = re.sub('\n{1,}', '\n', text)
|
|
| 202 |
splitText = text.split('\n')
|
|
| 203 |
textGroupIndex = 0 |
|
| 204 |
|
|
| 205 |
lastRT = (-1, -1) |
|
| 206 |
tempText = '' |
|
| 207 |
charWidth = -1 |
|
| 208 |
charHeight = -1 |
|
| 209 |
lineSp = (-1, -1) |
|
| 210 |
lineRect = None |
|
| 211 |
for index in range(len(boundaryOcrData.split('\n'))):
|
|
| 212 |
data = boundaryOcrData.split('\n')[index]
|
|
| 213 |
sData = data.split(' ')
|
|
| 214 |
if len(sData) >= 5: |
|
| 215 |
char = sData[0] |
|
| 216 |
csx = int(sData[1]) |
|
| 217 |
csy = imgHeight - int(sData[4]) |
|
| 218 |
cex = int(sData[3]) |
|
| 219 |
cey = imgHeight - int(sData[2]) |
|
| 220 |
cw = cex - csx |
|
| 221 |
ch = abs(cey - csy) |
|
| 222 |
charWidth = max(charWidth, cw) |
|
| 223 |
charHeight = max(charHeight, ch) |
|
| 224 |
currentRect = None |
|
| 225 |
if lastRT != (-1, -1): |
|
| 226 |
currentRect = QRect(csx, csy, csx + cw, csy + ch) |
|
| 227 |
if lastRT == (-1, -1) and lineRect is None: |
|
| 228 |
tempText = tempText + char |
|
| 229 |
lastRT = (csx+cw, csy) |
|
| 230 |
lineSp = (csx, csy) |
|
| 231 |
lineRect = QRect(lineSp[0], lineSp[1], imgWidth - lineSp[0], charHeight) |
|
| 232 |
elif (lineRect is not None and currentRect is not None) and lineRect.intersects(currentRect): |
|
| 233 |
tempText = tempText + char |
|
| 234 |
lastRT = (csx+cw, min(lineSp[1], csy)) |
|
| 235 |
lineSp = (lineSp[0], min(lineSp[1], csy)) |
|
| 236 |
lineHeight = max(max(lineRect.height(), abs(cey - lineSp[1])), abs((lineRect.y()+lineRect.height()) - lineSp[1])) |
|
| 237 |
lineRect.setX(lineSp[0]) |
|
| 238 |
lineRect.setY(lineSp[1]) |
|
| 239 |
lineRect.setHeight(lineHeight) |
|
| 198 |
boundaryOcrData = pytesseract.image_to_boxes(im, config=conf, lang='eng') |
|
| 199 |
bounding_boxes = boundaryOcrData.split('\n')
|
|
| 200 |
merged_boxes = [] |
|
| 201 |
for box in bounding_boxes: |
|
| 202 |
if merged_boxes: |
|
| 203 |
tokens = box.split(' ')
|
|
| 204 |
if len(tokens) >= 5: |
|
| 205 |
minx = int(tokens[1]) |
|
| 206 |
miny = int(tokens[2]) |
|
| 207 |
maxx = int(tokens[3]) |
|
| 208 |
maxy = int(tokens[4]) |
|
| 209 |
|
|
| 210 |
top = merged_boxes[-1].top() |
|
| 211 |
bottom = merged_boxes[-1].bottom() |
|
| 212 |
|
|
| 213 |
if maxy < top or miny > bottom: |
|
| 214 |
merged_boxes.append(QRect(minx, miny, maxx - minx, maxy - miny)) |
|
| 240 | 215 |
else: |
| 241 |
# Save previous line |
|
| 242 |
if len(splitText) > textGroupIndex: |
|
| 243 |
rect = QRect(lineSp[0], lineSp[1], lastRT[0]-lineSp[0], lineRect.height()) |
|
| 244 |
if rect.height() >= minSize and rect.height() <= maxSize: |
|
| 245 |
if angle == 90 or angle == 270: |
|
| 246 |
transform = QTransform() |
|
| 247 |
transform.translate(imgHeight*0.5, imgWidth*0.5) |
|
| 248 |
transform.rotate(-angle) |
|
| 249 |
transform.translate(-imgWidth*0.5, -imgHeight*0.5) |
|
| 250 |
rect = transform.mapRect(rect) |
|
| 251 |
prevLineText = ti.TextInfo(splitText[textGroupIndex], startPoint[0]+rect.left(), startPoint[1]+rect.top(), rect.width(), rect.height(), angle) |
|
| 252 |
textInfoList.append(prevLineText) |
|
| 253 |
textGroupIndex = textGroupIndex + 1 |
|
| 254 |
|
|
| 255 |
# Start new line |
|
| 256 |
tempText = char |
|
| 257 |
charWidth = cw |
|
| 258 |
charHeight = ch |
|
| 259 |
lastRT = (csx + cw, csy) |
|
| 260 |
lineSp = (csx, csy) |
|
| 261 |
lineRect = QRect(lineSp[0], lineSp[1], imgWidth - lineSp[0], ch) |
|
| 262 |
|
|
| 263 |
if not (len(textInfoList) > textGroupIndex) and textGroupIndex < len(splitText): |
|
| 264 |
rect = QRect(lineSp[0], lineSp[1], lastRT[0]-lineSp[0], lineRect.height()) |
|
| 265 |
if rect.height() >= minSize and rect.height() <= maxSize: |
|
| 266 |
if angle == 90 or angle == 270: |
|
| 267 |
transform = QTransform() |
|
| 268 |
transform.translate(imgHeight*0.5, imgWidth*0.5) |
|
| 269 |
transform.rotate(-angle) |
|
| 270 |
transform.translate(-imgWidth*0.5, -imgHeight*0.5) |
|
| 271 |
rect = transform.mapRect(rect) |
|
| 272 |
textInfo = ti.TextInfo(splitText[textGroupIndex], startPoint[0]+rect.left(), startPoint[1]+rect.top(), rect.width(), rect.height(), angle) |
|
| 273 |
if textInfo not in textInfoList: |
|
| 274 |
textInfoList.append(textInfo) |
|
| 216 |
merged_boxes[-1].setLeft(min(merged_boxes[-1].left(), minx)) |
|
| 217 |
merged_boxes[-1].setTop(min(merged_boxes[-1].top(), miny)) |
|
| 218 |
merged_boxes[-1].setRight(max(merged_boxes[-1].right(), maxx)) |
|
| 219 |
merged_boxes[-1].setBottom(max(merged_boxes[-1].bottom(), maxy)) |
|
| 275 | 220 |
else: |
| 276 |
return None |
|
| 277 |
else: |
|
| 278 |
pass |
|
| 221 |
tokens = box.split(' ')
|
|
| 222 |
if len(tokens) >= 5: |
|
| 223 |
minx = int(tokens[1]) |
|
| 224 |
miny = int(tokens[2]) |
|
| 225 |
maxx = int(tokens[3]) |
|
| 226 |
maxy = int(tokens[4]) |
|
| 227 |
merged_boxes.append(QRect(minx, miny, maxx - minx, maxy - miny)) |
|
| 228 |
|
|
| 229 |
for rect in merged_boxes: |
|
| 230 |
cropped = im.crop((rect.left(), imgHeight - rect.bottom(), rect.right(), imgHeight - rect.top())) |
|
| 231 |
text = pytesseract.image_to_string(cropped, config=conf, lang='eng') |
|
| 232 |
|
|
| 233 |
if rect.height() >= minSize and rect.height() <= maxSize: |
|
| 234 |
text_rect = QRect(rect.left(), imgHeight - rect.bottom(), rect.width(), rect.height()) |
|
| 235 |
if angle == 90 or angle == 270: |
|
| 236 |
transform = QTransform() |
|
| 237 |
transform.translate(imgHeight*0.5, imgWidth*0.5) |
|
| 238 |
transform.rotate(-angle) |
|
| 239 |
transform.translate(-imgWidth*0.5, -imgHeight*0.5) |
|
| 240 |
text_rect = transform.mapRect(text_rect) |
|
| 241 |
|
|
| 242 |
text_info = ti.TextInfo(text, startPoint[0] + text_rect.left(), startPoint[1] + text_rect.top(), text_rect.width(), text_rect.height(), angle) |
|
| 243 |
textInfoList.append(text_info) |
|
| 244 |
|
|
| 279 | 245 |
return textInfoList |
| 280 | 246 |
except Exception as ex: |
| 281 | 247 |
from App import App |
내보내기 Unified diff
