| commit | author | age | ||
| e555c0 | 1 | from pyzbar.pyzbar import decode |
| 0d97e9 | 2 | from .sid_process import getSID |
| e555c0 | 3 | import cv2 |
| SP | 4 | import numpy as np |
| 69abed | 5 | import os |
| 0d97e9 | 6 | import pkg_resources |
| SP | 7 | |
| d88ce4 | 8 | markerfile = '/template-sq.png' # always use slash |
| 0d97e9 | 9 | markerfilename = pkg_resources.resource_filename(__name__, markerfile) |
| SP | 10 | |
| e555c0 | 11 | |
| SP | 12 | |
| 511c2e | 13 | class Paper: |
| 69abed | 14 | def __init__(self, filename=None, sid_classifier=None, settings=None, output_path="/tmp"): |
| 511c2e | 15 | self.filename = filename |
| 69abed | 16 | self.output_path=output_path |
| 511c2e | 17 | self.invalid = None |
| SP | 18 | self.QRData = None |
| e0996e | 19 | self.settings = {"answer_threshold": 0.25} if settings is None else settings |
| 511c2e | 20 | self.errors = [] |
| SP | 21 | self.warnings = [] |
| e0996e | 22 | self.sid = None |
| 762a5e | 23 | self.sid_classifier = sid_classifier |
| 511c2e | 24 | if filename is not None: |
| SP | 25 | self.loadImage(filename) |
| 26 | self.runOcr() | |
| e555c0 | 27 | |
| 511c2e | 28 | def loadImage(self, filename, rgbchannel=0): |
| SP | 29 | self.img = cv2.imread(filename, rgbchannel) |
| 30 | if self.img is None: | |
| 31 | self.errors.append("File could not be loaded!") | |
| 32 | self.invalid = True | |
| 33 | return | |
| 34 | self.imgHeight, self.imgWidth = self.img.shape[0:2] | |
| e555c0 | 35 | |
| 0d97e9 | 36 | def saveImage(self, filename="/tmp/debug_image.png"): |
| 511c2e | 37 | cv2.imwrite(filename, self.img) |
| e555c0 | 38 | |
| 511c2e | 39 | def runOcr(self): |
| SP | 40 | if self.invalid == True: |
| 41 | return | |
| 42 | self.decodeQRandRotate() | |
| 43 | self.imgTreshold() | |
| d88ce4 | 44 | cv2.imwrite('/tmp/debug_threshold.png', self.bwimg) |
| 511c2e | 45 | skewAngle = 0 |
| SP | 46 | # try: |
| 47 | # skewAngle=self.getSkewAngle() | |
| 48 | # except: | |
| 49 | # self.errors.append("Could not determine skew angle!") | |
| 50 | # self.rotateAngle(skewAngle) | |
| e555c0 | 51 | |
| 511c2e | 52 | self.generateAnswerMatrix() |
| e555c0 | 53 | |
| 511c2e | 54 | self.saveImage() |
| e555c0 | 55 | |
| 511c2e | 56 | def decodeQRandRotate(self): |
| SP | 57 | if self.invalid == True: |
| 58 | return | |
| 59 | blur = cv2.blur(self.img, (3, 3)) | |
| 60 | d = decode(blur) | |
| 61 | self.img = blur | |
| 62 | if len(d) == 0: | |
| 63 | self.errors.append("QR code could not be found!") | |
| 64 | self.data = None | |
| 65 | self.invalid = True | |
| 66 | return | |
| e0996e | 67 | if(len(d)>1): #if there are multiple codes, get first ean or qr code available. |
| SP | 68 | for dd in d: |
| 69 | if(dd.type=="EAN13" or dd.type=="QR"): | |
| 70 | d[0]=dd | |
| 71 | break | |
| 511c2e | 72 | self.QRDecode = d |
| SP | 73 | self.QRData = d[0].data |
| 74 | xpos = d[0].rect.left | |
| 75 | ypos = d[0].rect.top | |
| 76 | # check if image is rotated wrongly | |
| 82ec6d | 77 | if xpos > self.imgHeight / 2.0 and ypos > self.imgWidth / 2.0: |
| 511c2e | 78 | self.rotateAngle(180) |
| e555c0 | 79 | |
| 511c2e | 80 | def rotateAngle(self, angle=0): |
| e0996e | 81 | # rot_mat = cv2.getRotationMatrix2D( |
| 82ec6d | 82 | # (self.imgHeight / 2, self.imgWidth / 2), angle, 1.0 |
| e0996e | 83 | # ) |
| 511c2e | 84 | rot_mat = cv2.getRotationMatrix2D( |
| e0996e | 85 | (self.imgWidth / 2, self.imgHeight / 2), angle, 1.0 |
| 511c2e | 86 | ) |
| SP | 87 | result = cv2.warpAffine( |
| 88 | self.img, | |
| 89 | rot_mat, | |
| 82ec6d | 90 | (self.imgWidth, self.imgHeight), |
| 511c2e | 91 | flags=cv2.INTER_CUBIC, |
| SP | 92 | borderMode=cv2.BORDER_CONSTANT, |
| 93 | borderValue=(255, 255, 255), | |
| 94 | ) | |
| e555c0 | 95 | |
| 511c2e | 96 | self.img = result |
| SP | 97 | self.imgHeight, self.imgWidth = self.img.shape[0:2] |
| e555c0 | 98 | |
| 511c2e | 99 | # todo, make better tresholding |
| SP | 100 | def imgTreshold(self): |
| 101 | (self.thresh, self.bwimg) = cv2.threshold( | |
| 102 | self.img, 128, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU | |
| 103 | ) | |
| e555c0 | 104 | |
| 511c2e | 105 | def getSkewAngle(self): |
| SP | 106 | neg = 255 - self.bwimg # get negative image |
| 0d97e9 | 107 | cv2.imwrite("/tmp/debug_1.png", neg) |
| e555c0 | 108 | |
| 511c2e | 109 | angle_counter = 0 # number of angles |
| SP | 110 | angle = 0.0 # collects sum of angles |
| 111 | cimg = cv2.cvtColor(self.img, cv2.COLOR_GRAY2BGR) | |
| e555c0 | 112 | |
| 511c2e | 113 | # get all the Hough lines |
| SP | 114 | for line in cv2.HoughLinesP(neg, 1, np.pi / 180, 325): |
| 115 | x1, y1, x2, y2 = line[0] | |
| 116 | cv2.line(cimg, (x1, y1), (x2, y2), (0, 0, 255), 2) | |
| 117 | # calculate the angle (in radians) | |
| 118 | this_angle = np.arctan2(y2 - y1, x2 - x1) | |
| 119 | if this_angle and abs(this_angle) <= 10: | |
| 120 | # filtered zero degree and outliers | |
| 121 | angle += this_angle | |
| 122 | angle_counter += 1 | |
| e555c0 | 123 | |
| 511c2e | 124 | # the skew is calculated of the mean of the total angles, #try block helps with division by zero. |
| SP | 125 | try: |
| 126 | skew = np.rad2deg( | |
| 127 | angle / angle_counter | |
| 128 | ) # the 1.2 factor is just experimental.... | |
| 129 | except: | |
| 130 | skew = 0 | |
| e555c0 | 131 | |
| 0d97e9 | 132 | cv2.imwrite("/tmp/debug_2.png", cimg) |
| 511c2e | 133 | return skew |
| e555c0 | 134 | |
| 511c2e | 135 | def locateUpMarkers(self, threshold=0.85, height=200): |
| 0d97e9 | 136 | template = cv2.imread(markerfilename, 0) |
| 511c2e | 137 | w, h = template.shape[::-1] |
| d88ce4 | 138 | crop_img = self.bwimg[0:height, :] |
| 511c2e | 139 | res = cv2.matchTemplate(crop_img, template, cv2.TM_CCOEFF_NORMED) |
| SP | 140 | loc = np.where(res >= threshold) |
| 141 | cimg = cv2.cvtColor(crop_img, cv2.COLOR_GRAY2BGR) | |
| 142 | # remove false matching of the squares in qr code | |
| 143 | loc_filtered_x = [] | |
| 144 | loc_filtered_y = [] | |
| 145 | if len(loc[0]) == 0: | |
| 146 | min_y = -1 | |
| 147 | else: | |
| 148 | min_y = np.min(loc[0]) | |
| 149 | for pt in zip(*loc[::-1]): | |
| 150 | if pt[1] < min_y + 20: | |
| 151 | loc_filtered_y.append(pt[1]) | |
| 152 | loc_filtered_x.append(pt[0]) | |
| 153 | # order by x coordinate | |
| 154 | loc_filtered_x, loc_filtered_y = zip( | |
| 155 | *sorted(zip(loc_filtered_x, loc_filtered_y)) | |
| 156 | ) | |
| 02e0f7 | 157 | # loc=[loc_filtered_y,loc_filtered_x] |
| SP | 158 | # remove duplicates |
| 511c2e | 159 | a = np.diff(loc_filtered_x) > 40 |
| SP | 160 | a = np.append(a, True) |
| 161 | loc_filtered_x = np.array(loc_filtered_x) | |
| 162 | loc_filtered_y = np.array(loc_filtered_y) | |
| 163 | loc = [loc_filtered_y[a], loc_filtered_x[a]] | |
| 164 | for pt in zip(*loc[::-1]): | |
| 165 | cv2.rectangle(cimg, pt, (pt[0] + w, pt[1] + h), (0, 255, 255), 2) | |
| e555c0 | 166 | |
| 0d97e9 | 167 | cv2.imwrite("/tmp/debug_3.png", cimg) |
| e555c0 | 168 | |
| 511c2e | 169 | self.xMarkerLocations = loc |
| SP | 170 | return loc |
| e555c0 | 171 | |
| 511c2e | 172 | def locateRightMarkers(self, threshold=0.85, width=200): |
| 0d97e9 | 173 | template = cv2.imread(markerfilename, 0) |
| 511c2e | 174 | w, h = template.shape[::-1] |
| d88ce4 | 175 | crop_img = self.bwimg[:, -width:] |
| SP | 176 | cv2.imwrite('/tmp/debug_right.png', crop_img) |
| 511c2e | 177 | res = cv2.matchTemplate(crop_img, template, cv2.TM_CCOEFF_NORMED) |
| SP | 178 | loc = np.where(res >= threshold) |
| 179 | cimg = cv2.cvtColor(crop_img, cv2.COLOR_GRAY2BGR) | |
| 180 | # remove false matching of the squares in qr code | |
| 181 | loc_filtered_x = [] | |
| 182 | loc_filtered_y = [] | |
| 183 | if len(loc[1]) == 0: | |
| 184 | min_x = -1 | |
| 185 | else: | |
| 186 | max_x = np.max(loc[1]) | |
| 187 | for pt in zip(*loc[::-1]): | |
| 188 | if pt[1] > max_x - 20: | |
| 189 | loc_filtered_y.append(pt[1]) | |
| 190 | loc_filtered_x.append(pt[0]) | |
| 191 | # order by y coordinate | |
| d88ce4 | 192 | try: |
| SP | 193 | loc_filtered_y, loc_filtered_x = zip( |
| 194 | *sorted(zip(loc_filtered_y, loc_filtered_x)) | |
| 195 | ) | |
| 196 | except: | |
| 197 | self.yMarkerLocations=[np.array([1,1]),np.array([1,2])] | |
| 198 | return self.yMarkerLocations | |
| 511c2e | 199 | # loc=[loc_filtered_y,loc_filtered_x] |
| SP | 200 | # remove duplicates |
| 201 | a = np.diff(loc_filtered_y) > 40 | |
| 202 | a = np.append(a, True) | |
| 203 | loc_filtered_x = np.array(loc_filtered_x) | |
| 204 | loc_filtered_y = np.array(loc_filtered_y) | |
| 205 | loc = [loc_filtered_y[a], loc_filtered_x[a]] | |
| 206 | for pt in zip(*loc[::-1]): | |
| 207 | cv2.rectangle(cimg, pt, (pt[0] + w, pt[1] + h), (0, 255, 255), 2) | |
| e555c0 | 208 | |
| 0d97e9 | 209 | cv2.imwrite("/tmp/debug_4.png", cimg) |
| e555c0 | 210 | |
| 511c2e | 211 | self.yMarkerLocations = [loc[0], loc[1] + self.imgWidth - width] |
| SP | 212 | return self.yMarkerLocations |
| e555c0 | 213 | |
| 511c2e | 214 | def generateAnswerMatrix(self): |
| SP | 215 | self.locateUpMarkers() |
| 216 | self.locateRightMarkers() | |
| e555c0 | 217 | |
| 511c2e | 218 | roixoff = 10 |
| SP | 219 | roiyoff = 5 |
| 220 | roiwidth = 50 | |
| 221 | roiheight = roiwidth | |
| 222 | totpx = roiwidth * roiheight | |
| e555c0 | 223 | |
| 511c2e | 224 | self.answerMatrix = [] |
| SP | 225 | for y in self.yMarkerLocations[0]: |
| 226 | oneline = [] | |
| 227 | for x in self.xMarkerLocations[1]: | |
| 228 | roi = self.bwimg[ | |
| 229 | y - roiyoff : y + int(roiheight - roiyoff), | |
| 230 | x - roixoff : x + int(roiwidth - roixoff), | |
| 231 | ] | |
| 232 | # cv2.imwrite('ans_x'+str(x)+'_y_'+str(y)+'.png',roi) | |
| 233 | black = totpx - cv2.countNonZero(roi) | |
| 234 | oneline.append(black / totpx) | |
| 235 | self.answerMatrix.append(oneline) | |
| 9efc18 | 236 | |
| SP | 237 | def get_enhanced_sid(self): |
| 02e0f7 | 238 | if self.sid_classifier is None: |
| SP | 239 | return "x" |
| 762a5e | 240 | if self.settings is not None: |
| e0996e | 241 | sid_mask = self.settings.get("sid_mask", None) |
| SP | 242 | es, err, warn = getSID( |
| 02e0f7 | 243 | self.img[ |
| d5c694 | 244 | int(0.04 * self.imgHeight) : int(0.095 * self.imgHeight), |
| 02e0f7 | 245 | int(0.7 * self.imgWidth) : int(0.99 * self.imgWidth), |
| SP | 246 | ], |
| 247 | self.sid_classifier, | |
| e0996e | 248 | sid_mask, |
| 02e0f7 | 249 | ) |
| 5cb7c1 | 250 | [self.errors.append(e) for e in err] |
| SP | 251 | [self.warnings.append(w) for w in warn] |
| 02e0f7 | 252 | return es |
| 0436f6 | 253 | |
| SP | 254 | def get_code_data(self): |
| cf921b | 255 | if self.QRData is None: |
| SP | 256 | self.errors.append("Could not read QR or EAN code! Not an exam?") |
| e0996e | 257 | retval = { |
| SP | 258 | "exam_id": None, |
| 259 | "page_no": None, | |
| 260 | "paper_id": None, | |
| 261 | "faculty_id": None, | |
| 262 | "sid": None, | |
| 0436f6 | 263 | } |
| e0996e | 264 | return retval |
| SP | 265 | qrdata = bytes.decode(self.QRData, "utf8") |
| 266 | if self.QRDecode[0].type == "EAN13": | |
| 267 | return { | |
| 268 | "exam_id": int(qrdata[0:7]), | |
| 69abed | 269 | "page_no": int(qrdata[7])+1, |
| e0996e | 270 | "paper_id": int(qrdata[-5:-1]), |
| SP | 271 | "faculty_id": None, |
| 272 | "sid": None, | |
| 273 | } | |
| 274 | else: | |
| 275 | data = qrdata.split(",") | |
| 276 | retval = { | |
| 277 | "exam_id": int(data[1]), | |
| 69abed | 278 | "page_no": int(data[3])+1, |
| e0996e | 279 | "paper_id": int(data[2]), |
| SP | 280 | "faculty_id": int(data[0]), |
| d88ce4 | 281 | "sid": None |
| e0996e | 282 | } |
| SP | 283 | if len(data) > 4: |
| 284 | retval["sid"] = data[4] | |
| 0436f6 | 285 | |
| SP | 286 | return retval |
| 287 | ||
| 288 | def get_paper_ocr_data(self): | |
| e0996e | 289 | data = self.get_code_data() |
| 69abed | 290 | data["qr"] = bytes.decode(self.QRData, 'utf8') |
| e0996e | 291 | data["errors"] = self.errors |
| SP | 292 | data["warnings"] = self.warnings |
| 293 | data["up_position"] = ( | |
| 294 | list(self.xMarkerLocations[1] / self.imgWidth), | |
| 295 | list(self.yMarkerLocations[1] / self.imgHeight), | |
| 296 | ) | |
| 297 | data["right_position"] = ( | |
| 298 | list(self.xMarkerLocations[1] / self.imgWidth), | |
| 299 | list(self.yMarkerLocations[1] / self.imgHeight), | |
| 300 | ) | |
| 301 | data["ans_matrix"] = ( | |
| 302 | (np.array(self.answerMatrix) > self.settings["answer_threshold"]) * 1 | |
| 303 | ).tolist() | |
| d88ce4 | 304 | if data["sid"] is None and data["page_no"] == 2: |
| e0996e | 305 | data["sid"] = self.get_enhanced_sid() |
| 69abed | 306 | output_filename=os.path.join(self.output_path, '.'.join(self.filename.split('/')[-1].split('.')[:-1])+".png") |
| SP | 307 | cv2.imwrite(output_filename, self.img) |
| 308 | data['output_filename']=output_filename | |
| 0436f6 | 309 | return data |
