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