from pyzbar.pyzbar import decode
|
from sid_process import getSID
|
import cv2
|
import numpy as np
|
import math
|
|
|
class Paper:
|
def __init__(self, filename=None, sid_classifier=None, settings=None):
|
self.filename = filename
|
self.invalid = None
|
self.QRData = None
|
self.settings={'answer_treshold':0.25,} if settings is None else settings
|
self.errors = []
|
self.warnings = []
|
self.sid=None
|
self.sid_classifier = sid_classifier
|
if filename is not None:
|
self.loadImage(filename)
|
self.runOcr()
|
|
def loadImage(self, filename, rgbchannel=0):
|
self.img = cv2.imread(filename, rgbchannel)
|
if self.img is None:
|
self.errors.append("File could not be loaded!")
|
self.invalid = True
|
return
|
self.imgHeight, self.imgWidth = self.img.shape[0:2]
|
|
def saveImage(self, filename="debug_image.png"):
|
cv2.imwrite(filename, self.img)
|
|
def runOcr(self):
|
if self.invalid == True:
|
return
|
self.decodeQRandRotate()
|
self.imgTreshold()
|
skewAngle = 0
|
# try:
|
# skewAngle=self.getSkewAngle()
|
# except:
|
# self.errors.append("Could not determine skew angle!")
|
# self.rotateAngle(skewAngle)
|
|
self.generateAnswerMatrix()
|
|
self.saveImage()
|
|
def decodeQRandRotate(self):
|
if self.invalid == True:
|
return
|
blur = cv2.blur(self.img, (3, 3))
|
d = decode(blur)
|
self.img = blur
|
if len(d) == 0:
|
self.errors.append("QR code could not be found!")
|
self.data = None
|
self.invalid = True
|
return
|
self.QRDecode = d
|
self.QRData = d[0].data
|
xpos = d[0].rect.left
|
ypos = d[0].rect.top
|
# check if image is rotated wrongly
|
if xpos > self.imgHeight / 2.0 and ypost > self.imgWidth / 2.0:
|
self.rotateAngle(180)
|
|
def rotateAngle(self, angle=0):
|
rot_mat = cv2.getRotationMatrix2D(
|
(self.imgHeight / 2, self.imgWidth / 2), angle, 1.0
|
)
|
result = cv2.warpAffine(
|
self.img,
|
rot_mat,
|
(self.imgHeight, self.imgWidth),
|
flags=cv2.INTER_CUBIC,
|
borderMode=cv2.BORDER_CONSTANT,
|
borderValue=(255, 255, 255),
|
)
|
|
self.img = result
|
self.imgHeight, self.imgWidth = self.img.shape[0:2]
|
|
# todo, make better tresholding
|
|
def imgTreshold(self):
|
(self.thresh, self.bwimg) = cv2.threshold(
|
self.img, 128, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU
|
)
|
|
def getSkewAngle(self):
|
neg = 255 - self.bwimg # get negative image
|
cv2.imwrite("debug_1.png", neg)
|
|
angle_counter = 0 # number of angles
|
angle = 0.0 # collects sum of angles
|
cimg = cv2.cvtColor(self.img, cv2.COLOR_GRAY2BGR)
|
|
# get all the Hough lines
|
for line in cv2.HoughLinesP(neg, 1, np.pi / 180, 325):
|
x1, y1, x2, y2 = line[0]
|
cv2.line(cimg, (x1, y1), (x2, y2), (0, 0, 255), 2)
|
# calculate the angle (in radians)
|
this_angle = np.arctan2(y2 - y1, x2 - x1)
|
if this_angle and abs(this_angle) <= 10:
|
# filtered zero degree and outliers
|
angle += this_angle
|
angle_counter += 1
|
|
# the skew is calculated of the mean of the total angles, #try block helps with division by zero.
|
try:
|
skew = np.rad2deg(
|
angle / angle_counter
|
) # the 1.2 factor is just experimental....
|
except:
|
skew = 0
|
|
cv2.imwrite("debug_2.png", cimg)
|
return skew
|
|
def locateUpMarkers(self, threshold=0.85, height=200):
|
template = cv2.imread("template.png", 0)
|
w, h = template.shape[::-1]
|
crop_img = self.img[0:height, :]
|
res = cv2.matchTemplate(crop_img, template, cv2.TM_CCOEFF_NORMED)
|
loc = np.where(res >= threshold)
|
cimg = cv2.cvtColor(crop_img, cv2.COLOR_GRAY2BGR)
|
# remove false matching of the squares in qr code
|
loc_filtered_x = []
|
loc_filtered_y = []
|
if len(loc[0]) == 0:
|
min_y = -1
|
else:
|
min_y = np.min(loc[0])
|
for pt in zip(*loc[::-1]):
|
if pt[1] < min_y + 20:
|
loc_filtered_y.append(pt[1])
|
loc_filtered_x.append(pt[0])
|
# order by x coordinate
|
loc_filtered_x, loc_filtered_y = zip(
|
*sorted(zip(loc_filtered_x, loc_filtered_y))
|
)
|
# loc=[loc_filtered_y,loc_filtered_x]
|
# remove duplicates
|
a = np.diff(loc_filtered_x) > 40
|
a = np.append(a, True)
|
loc_filtered_x = np.array(loc_filtered_x)
|
loc_filtered_y = np.array(loc_filtered_y)
|
loc = [loc_filtered_y[a], loc_filtered_x[a]]
|
for pt in zip(*loc[::-1]):
|
cv2.rectangle(cimg, pt, (pt[0] + w, pt[1] + h), (0, 255, 255), 2)
|
|
cv2.imwrite("debug_3.png", cimg)
|
|
self.xMarkerLocations = loc
|
return loc
|
|
def locateRightMarkers(self, threshold=0.85, width=200):
|
template = cv2.imread("template.png", 0)
|
w, h = template.shape[::-1]
|
crop_img = self.img[:, -width:]
|
res = cv2.matchTemplate(crop_img, template, cv2.TM_CCOEFF_NORMED)
|
loc = np.where(res >= threshold)
|
cimg = cv2.cvtColor(crop_img, cv2.COLOR_GRAY2BGR)
|
# remove false matching of the squares in qr code
|
loc_filtered_x = []
|
loc_filtered_y = []
|
if len(loc[1]) == 0:
|
min_x = -1
|
else:
|
max_x = np.max(loc[1])
|
for pt in zip(*loc[::-1]):
|
if pt[1] > max_x - 20:
|
loc_filtered_y.append(pt[1])
|
loc_filtered_x.append(pt[0])
|
# order by y coordinate
|
loc_filtered_y, loc_filtered_x = zip(
|
*sorted(zip(loc_filtered_y, loc_filtered_x))
|
)
|
# loc=[loc_filtered_y,loc_filtered_x]
|
# remove duplicates
|
a = np.diff(loc_filtered_y) > 40
|
a = np.append(a, True)
|
loc_filtered_x = np.array(loc_filtered_x)
|
loc_filtered_y = np.array(loc_filtered_y)
|
loc = [loc_filtered_y[a], loc_filtered_x[a]]
|
for pt in zip(*loc[::-1]):
|
cv2.rectangle(cimg, pt, (pt[0] + w, pt[1] + h), (0, 255, 255), 2)
|
|
cv2.imwrite("debug_4.png", cimg)
|
|
self.yMarkerLocations = [loc[0], loc[1] + self.imgWidth - width]
|
return self.yMarkerLocations
|
|
def generateAnswerMatrix(self):
|
self.locateUpMarkers()
|
self.locateRightMarkers()
|
|
roixoff = 10
|
roiyoff = 5
|
roiwidth = 50
|
roiheight = roiwidth
|
totpx = roiwidth * roiheight
|
|
self.answerMatrix = []
|
for y in self.yMarkerLocations[0]:
|
oneline = []
|
for x in self.xMarkerLocations[1]:
|
roi = self.bwimg[
|
y - roiyoff : y + int(roiheight - roiyoff),
|
x - roixoff : x + int(roiwidth - roixoff),
|
]
|
# cv2.imwrite('ans_x'+str(x)+'_y_'+str(y)+'.png',roi)
|
black = totpx - cv2.countNonZero(roi)
|
oneline.append(black / totpx)
|
self.answerMatrix.append(oneline)
|
|
def get_enhanced_sid(self):
|
if self.sid_classifier is None:
|
return "x"
|
if self.settings is not None:
|
sid_mask=self.settings.get("sid_mask", None)
|
es,err,warn = getSID(
|
self.img[
|
int(0.04 * self.imgHeight) : int(0.095 * self.imgHeight),
|
int(0.7 * self.imgWidth) : int(0.99 * self.imgWidth),
|
],
|
self.sid_classifier,
|
sid_mask
|
)
|
[self.errors.append(e) for e in err]
|
[self.warnings.append(w) for w in warn]
|
return es
|
|
|
def get_code_data(self):
|
qrdata = bytes.decode(self.QRData, 'utf8')
|
if self.QRDecode[0].type=='EAN13':
|
return {'exam_id': int(qrdata[0:7]),
|
'page_no': int(qrdata[7]),
|
'paper_id': int(qrdata[-5:-1]),
|
'faculty_id': None,
|
'sid': None
|
}
|
else:
|
data=qrdata.split(',')
|
retval={'exam_id': int(data[1]),
|
'page_no': int(data[3]),
|
'paper_id':int(data[2]),
|
'faculty_id':int(data[0]),
|
}
|
if(len(data)>4):
|
retval['sid']=data[4]
|
|
return retval
|
|
def get_paper_ocr_data(self):
|
data=self.get_code_data()
|
data['qr']=self.QRData
|
data['errors']=self.errors
|
data['warnings']=self.warnings
|
data['up_position']=(list(self.xMarkerLocations[1]/self.imgWidth), list(self.yMarkerLocations[1]/self.imgHeight))
|
data['right_position']=(list(self.xMarkerLocations[1]/self.imgWidth), list(self.yMarkerLocations[1]/self.imgHeight))
|
data['ans_matrix']=((np.array(self.answerMatrix)>self.settings['answer_treshold'])*1).tolist()
|
if data['sid'] is None:
|
data['sid']=self.get_enhanced_sid()
|
return data
|