def detect(self, img):
if img.shape[0] < self.winSize_[1] or img.shape[1] < self.winSize_[0]:
img = imageProcessor.resize(img, self.winSize_[0],
self.winSize_[1])
# detector = cv2.HOGDescriptor()
# detector.setSVMDetector( cv2.HOGDescriptor_getDefaultPeopleDetector() )
# detector.save("hog.xml")
detector = imageProcessor.getMultiScaleDetector(self.winSize_)
supportVector = self.classifier_.getSupportVectors()[0]
supportVector = self.__convert2ColSample(supportVector, "float32")
detector.setSVMDetector(supportVector)
img = imutils.resize(img, width=min(960, img.shape[1]))
pad = 15
rois, weights = detector.detectMultiScale(img,
winStride=(4, 4),
padding=(pad, pad),
scale=1.2)
# apply non-maxima suppression to the bounding boxes using a
# fairly large overlap threshold to try to maintain overlapping
# boxes that are still people
rois = np.array([[x, y, x + w, y + h] for (x, y, w, h) in rois])
rois = non_max_suppression(rois, probs=None, overlapThresh=0.65)
for (x1, y1, x2, y2) in rois:
cv2.rectangle(img, (x1, y1), (x2, y2), (0, 0, 255), 2)
return img, rois
def predict(self, img):
if img is None:
print "detector detect(): no image input"
quit()
# check the size of the training sample, if not equal to the windowSize then resize it
if img.shape[0] != self.winSize_[1] or img.shape[1] != self.winSize_[0]:
img = imageProcessor.resize(img, self.winSize_[0],
self.winSize_[1])
# get the feature from this img file, the output is an 2d np.array
feature = self.featureExtract(img)
# convert the feature into ROW_SAMPLE form for the training of classifier
# the output is a 1d np.array
rowData = self.__convert2RowSample(feature, "float32")
# convert the 1d np.array to a 2d np.array with only 1 row
rowData = np.array([rowData], dtype="float32")
# the return value is a 2d np.array, res[i][0] is the result for the i-th item
res = self.classifier_.predict(rowData)
return res[0][0]
def train(self, dataSet, labels):
featureVectorSet = []
for img in dataSet:
# check the size of the training sample, if not equal to the windowSize then resize it
if img.shape[0] != self.winSize_[1] or img.shape[
1] != self.winSize_[0]:
img = imageProcessor.resize(img, self.winSize_[0],
self.winSize_[1])
# get the feature from this img file, the output is an 2d np.array
feature = self.featureExtract(img)
# convert the feature into ROW_SAMPLE form for the training of classifier
# the output is a 1d np.array
feature = self.__convert2RowSample(feature, "float32")
featureVectorSet.append(feature)
# convert the vectorSet from a list of 1d np.array to a 2d np.array
featureVectorSet = np.array(featureVectorSet, dtype="float32")
# convert the list of labels to a 1d np.array
labels = np.array(labels, dtype="int32")
self.classifier_.train(featureVectorSet, labels)
end='\r')
im = imageProcessor.getImage(mpath, imgSize)
print(' Processing (' + str(path.getPercentage()) +
'% Completed) - {0} - Image Uploaded '.format(
path.getCurrentFileName()),
end='\r')
imageProcessor.sliceImg(im)
print(' Processing (' + str(path.getPercentage()) +
'% Completed) - {0} - Smart Cropping Completed '.format(
path.getCurrentFileName()),
end='\r')
imageProcessor.arr2rImage(im)
im.RealImage = imageProcessor.resize(im.RealImage,
(imgSize - (imgMargin * 2)))
im.RealImage = imageProcessor.expand(im.RealImage, imgSize, im.bgColor)
print(' Processing (' + str(path.getPercentage()) +
'% Completed) - {0} - Smart Resizing Completed '.format(
path.getCurrentFileName()),
end='\r')
imageProcessor.saveImage(im, path.getCurrentDestination(), imgQuality)
print(
' ',
end='\r')
print(' {0} Processed and saved as {1}'.format(
path.getCurrentFileName(), path.getCurrentDestination()))
if im.bgDifference > 50:
path.moveProcessed('error')
x, y, w, h = cv2.boundingRect(c)
if (pow(max(h, w), 2) > (imgHeight * imgWidth) / 1200
and pow(max(h, w), 2) < (imgHeight * imgWidth) / 5):
img = binImg[y:y + h, x:x + w]
char = Character(img, x, y, w, h)
characters.append(char)
imgWithBounds = ip.drawCharacterBounds(binImg.copy(), characters)
cv2.imwrite("figures/preCharProcessing.jpg", imgWithBounds)
cp.processCharacters(characters)
imgWithBounds = ip.drawCharacterBounds(binImg.copy(), characters)
cv2.imwrite("figures/postCharProcessing.jpg", imgWithBounds)
cv2.imwrite("figures/characterPreFormatting.jpg", characters[0].image)
for char in characters:
ip.squareImage(char)
ip.resize(char, (45, 45))
cv2.imwrite("figures/characterPostFormatting.jpg", characters[0].image)
classifiedImg = ip.drawClassifiedCharacters(imgWithBounds, classifier,
characters)
cv2.imwrite("figures/classifiedCharacters.jpg", classifiedImg)
lines = parseEquation(characters)
final = binImg.copy()
drawSolvedEquations(lines, final, 'x')
cv2.imwrite("figures/solvedEquation.jpg", final)