def five_Examples(test_data, prediction, target, path=file_handler.Path()):
count = 0
font_size = 50
for i in np.arange(len(test_data)):
if prediction[i] == target[i]:
# rc('text', usetex=True)
image_helpers.show_image(test_data[i])
plt.xlabel('Prediction: '+prediction[i], fontsize=font_size)
plt.savefig(path.figure + 'ex/p_{}.pdf'.format(count), format='pdf', dpi=1000)
plt.subplots_adjust(bottom=0.33)
plt.margins(0.2, 0)
plt.draw()
count += 1
if count > 4:
break
count = 0
for i in np.arange(len(test_data)):
if prediction[i] != target[i]:
# rc('text', usetex=True)
image_helpers.show_image(test_data[i])
plt.xlabel('Prediction: ' + prediction[i] + '\n Target: ' + target[i] , fontsize=font_size) # adjust buttom
# plt.xlabel('Prediction: ' + prediction[i] + ' | Target: ' + target[i], fontsize=font_size)
plt.subplots_adjust(bottom=0.33)
plt.savefig(path.figure + 'ex/n_{}.pdf'.format(count), format='pdf', dpi=1000)
plt.draw()
count += 1
if count > 4:
break
def classify(train_data,
train_target,
test_data,
test_target,
n_pca=49,
n_neighbors=4,
display=False):
test_data_original = test_data
train_data = pp.standardized_augmentation(train_data, display=False)
test_data = pp.standardized_augmentation(test_data, display=False)
train_data = pp.histogram_of_oriented_gradients(train_data)
if display:
plt.figure()
image_helpers.show_image(data[0])
plt.savefig(path.figure + 'pp/hog.pdf', format='pdf', dpi=1000)
plt.draw()
pca = PCA(n_components=n_pca)
pca_model = pca.fit(train_data)
pca_train = pca_model.transform(train_data)
pca_test = pca_model.transform(test_data)
knn_model = kNN.find_nearest_neighbor(pca_train,
train_target,
n_neighbors=n_neighbors)
prediction = kNN.predict(pca_test, knn_model)
error = calculate_error(prediction, test_target)
five_Examples(test_data_original, prediction, test_target)
return error
def character_detection(pca_model):
c = Colors()
path_detect_1 = '../database/detection-images/detection-1.jpg'
path_detect_2 = '../database/detection-images/detection-2.jpg'
image_1 = file_handler.load_image(path_detect_1)
image_1 = file_handler.normalize(image_1)
image_1_bool = image_1 < 0.99999
window_size = 20
non_overlap = 2
print('Shape of image: ', image_1.shape)
plt.figure()
visualize.show_image(image_1, colorbar=True)
for i in np.arange(0, len(image_1_bool)-window_size):
for j in np.arange(0, len(image_1_bool[i])-window_size):
window = image_1_bool[i:i+window_size, j:j+window_size]
window_sum = np.sum(window)
pca_val = pca_model.transform(np.array(window).flatten())
if pca_val > -5:
plt.scatter(j + 10, i + 10, c='r')
print('Window num ', i, j, '- Sum of cell: ', window_sum)
if window_sum > 190:
print('Window num ', i, j, '- Sum of cell: ', window_sum)
print(c.OKBLUE, 'PCA: ', pca_model.transform(np.array(window).flatten()), c.ENDC)
else:
print(c.ENDC, 'PCA: ', pca_model.transform(np.array(window).flatten()))
def standardized_augmentation(data, display=False):
path = Path()
if display:
plt.figure()
image_helpers.show_image(data[0])
plt.savefig(path.figure + 'pp/raw.pdf', format='pdf', dpi=1000)
plt.draw()
data_0 = image_helpers.convert_to_sensor_values(data)
if display:
plt.figure()
image_helpers.show_image(data_0[0])
plt.savefig(path.figure + 'pp/sensor.pdf', format='pdf', dpi=1000)
plt.draw()
data = denoise(data)
if display:
plt.figure()
image_helpers.show_image(data[0])
plt.savefig(path.figure + 'pp/denoise.pdf', format='pdf', dpi=1000)
plt.draw()
data = normalize(data)
data = image_to_bool(data)
if display:
plt.figure()
image_helpers.show_image(data[0])
plt.savefig(path.figure + 'pp/img_to_bool.pdf', format='pdf', dpi=1000)
plt.draw()
data = normalize(data)
return data