def getDataNonMalik(imageComplete):
# SHUFFLE THE IMAGES
random.shuffle(imageComplete)
# Arrays to hold the shuffled data and labels
shuffledData = []
shuffledLabels = []
for elem in imageComplete:
shuffledData.append((elem[0]).flatten()) # Use a simple array of pixels as the feature
shuffledLabels.append((elem[1][0]))
return shuffledData, shuffledLabels, imageComplete
def getDataNonMalik(imageComplete):
# SHUFFLE THE IMAGES
random.shuffle(imageComplete)
# Arrays to hold the shuffled data and labels
shuffledData = []
shuffledLabels = []
for elem in imageComplete:
shuffledData.append(
(elem[0]).flatten()) # Use a simple array of pixels as the feature
shuffledLabels.append((elem[1][0]))
return shuffledData, shuffledLabels, imageComplete
def getDataMALIK(gauss_bool, imageData):
#############
# Ink Normalization
#############
for i in range(len(imageData)):
aux_norm = np.linalg.norm(imageData[i])
if aux_norm != 0:
imageData[i] /= aux_norm
imageComplete = zip(imageData, imageLabels)
if DEBUG:
print 50*'-'
print ("Shapes of image data and labels: ", imageData.shape,
imageLabels.shape, len(imageComplete))
print "Image/Digit 10000:\n", imageComplete[20000]
############# SET ASIDE VALIDATION DATA (10,000) #############
# SHUFFLE THE IMAGES
random.shuffle(imageComplete)
shuffledData, shuffledLabels = getDataPickle()
if len(shuffledLabels)>0:
return shuffledData, shuffledLabels
n_bins=9
for ind in range(len(imageComplete)):
if ind % 100 ==0:
print 'feature extraction :' + str(ind*100./len(imageComplete))+ ' % over'
if gauss_bool:
gaussFirst_x = filters.gaussian_filter1d(imageComplete[i][0], 1, order = 1, axis = 0)
gaussFirst_y = filters.gaussian_filter1d(imageComplete[i][0], 1, order = 1, axis = 1)
ori = np.array(np.arctan2(gaussFirst_y, gaussFirst_x))
else:
grad_filter = np.array([[-1, 0, 1]])
gradx = signal.convolve2d(imageComplete[i][0], grad_filter, 'same')
grady = signal.convolve2d(imageComplete[i][0], np.transpose(grad_filter), 'same')
ori = np.array(np.arctan2(grady, gradx))
ori_4_hist = list()
ori_7_hist = list()
ori_4_1 = blockshaped(ori, 4, 4)
ori_4_2 = blockshaped(ori[2:-2, 2:-2], 4, 4)
for (elem1, elem2) in zip(ori_4_1, ori_4_2):
ori_4_hist.append(np.histogram(elem1.flatten(), n_bins, (-np.pi, np.pi))[0])
ori_4_hist.append(np.histogram(elem2.flatten(), n_bins, (-np.pi, np.pi))[0])
ori_7_1 = (blockshaped(ori, 7, 7))
ori_7_2 = (blockshaped(ori[3:-4, 3:-4], 7, 7))
for elem1, elem2 in zip(ori_7_1, ori_7_2):
ori_4_hist.append(np.histogram(elem1.flatten(), n_bins, (-np.pi, np.pi))[0])
ori_4_hist.append(np.histogram(elem2.flatten(), n_bins, (-np.pi, np.pi))[0])
ori_4_hist = np.float64(ori_4_hist)/(np.linalg.norm(ori_4_hist))
ori_7_hist = np.float64(ori_7_hist)/(np.linalg.norm(ori_7_hist))
shuffledData.append(np.append(ori_4_hist, ori_7_hist))
shuffledLabels.append((imageComplete[i][1][0]))
pickle.dump(shuffledData, open("./Results/shuffledData.p", 'wb'))
pickle.dump(shuffledLabels, open("./Results/shuffledLabels.p", 'wb'))
return shuffledData, shuffledLabels, imageComplete
def getDataMALIK(gauss_bool, imageData):
#############
# Ink Normalization
#############
for i in range(len(imageData)):
aux_norm = np.linalg.norm(imageData[i])
if aux_norm != 0:
imageData[i] /= aux_norm
imageComplete = zip(imageData, imageLabels)
if DEBUG:
print 50 * '-'
print("Shapes of image data and labels: ", imageData.shape,
imageLabels.shape, len(imageComplete))
print "Image/Digit 10000:\n", imageComplete[20000]
############# SET ASIDE VALIDATION DATA (10,000) #############
# SHUFFLE THE IMAGES
random.shuffle(imageComplete)
shuffledData, shuffledLabels = getDataPickle()
if len(shuffledLabels) > 0:
return shuffledData, shuffledLabels
n_bins = 9
for ind in range(len(imageComplete)):
if ind % 100 == 0:
print 'feature extraction :' + str(
ind * 100. / len(imageComplete)) + ' % over'
if gauss_bool:
gaussFirst_x = filters.gaussian_filter1d(imageComplete[i][0],
1,
order=1,
axis=0)
gaussFirst_y = filters.gaussian_filter1d(imageComplete[i][0],
1,
order=1,
axis=1)
ori = np.array(np.arctan2(gaussFirst_y, gaussFirst_x))
else:
grad_filter = np.array([[-1, 0, 1]])
gradx = signal.convolve2d(imageComplete[i][0], grad_filter, 'same')
grady = signal.convolve2d(imageComplete[i][0],
np.transpose(grad_filter), 'same')
ori = np.array(np.arctan2(grady, gradx))
ori_4_hist = list()
ori_7_hist = list()
ori_4_1 = blockshaped(ori, 4, 4)
ori_4_2 = blockshaped(ori[2:-2, 2:-2], 4, 4)
for (elem1, elem2) in zip(ori_4_1, ori_4_2):
ori_4_hist.append(
np.histogram(elem1.flatten(), n_bins, (-np.pi, np.pi))[0])
ori_4_hist.append(
np.histogram(elem2.flatten(), n_bins, (-np.pi, np.pi))[0])
ori_7_1 = (blockshaped(ori, 7, 7))
ori_7_2 = (blockshaped(ori[3:-4, 3:-4], 7, 7))
for elem1, elem2 in zip(ori_7_1, ori_7_2):
ori_4_hist.append(
np.histogram(elem1.flatten(), n_bins, (-np.pi, np.pi))[0])
ori_4_hist.append(
np.histogram(elem2.flatten(), n_bins, (-np.pi, np.pi))[0])
ori_4_hist = np.float64(ori_4_hist) / (np.linalg.norm(ori_4_hist))
ori_7_hist = np.float64(ori_7_hist) / (np.linalg.norm(ori_7_hist))
shuffledData.append(np.append(ori_4_hist, ori_7_hist))
shuffledLabels.append((imageComplete[i][1][0]))
pickle.dump(shuffledData, open("./Results/shuffledData.p", 'wb'))
pickle.dump(shuffledLabels, open("./Results/shuffledLabels.p", 'wb'))
return shuffledData, shuffledLabels, imageComplete
