def basic_hsv_mask(image):
"""
Mask based on low saturation and value (gray-black colors)
:param image: RGB numpy image
:return: image mask, True pixels are gray-black.
"""
hsv_image = rgb2hsv(image)
return np.bitwise_or(hsv_image[:, :, HSV_SAT_CHANNEL] <= MIN_SAT,
hsv_image[:, :, HSV_VAL_CHANNEL] <= MIN_VAL)
def hue_range_mask(image, min_hue, max_hue, sat_min=0.05):
hsv_image = rgb2hsv(image)
h_channel = gaussian(hsv_image[:, :, HSV_HUE_CHANNEL])
above_min = h_channel > min_hue
below_max = h_channel < max_hue
s_channel = gaussian(hsv_image[:, :, HSV_SAT_CHANNEL])
above_sat = s_channel > sat_min
return np.logical_and(np.logical_and(above_min, below_max), above_sat)
def test_overlay_custom_saturation():
rgb_img = np.random.uniform(size=(10, 10, 3))
labels = np.ones((10, 10), dtype=np.int64)
labels[5:, 5:] = 2
labels[:3, :3] = 0
alpha = 0.3
saturation = 0.3
rgb = label2rgb(labels,
image=rgb_img,
alpha=alpha,
bg_label=0,
saturation=saturation)
hsv = rgb2hsv(rgb_img)
hsv[..., 1] *= saturation
saturaded_img = hsv2rgb(hsv)
# check that rgb part of input image is saturated, where labels=0
assert_array_almost_equal(saturaded_img[:3, :3] * (1 - alpha), rgb[:3, :3])
color="white" if cm[i, j] > thresh else "black")
plt.tight_layout()
plt.ylabel('True label')
plt.xlabel('Predicted label')
plt.show()
#################### TRAINING ####################
# open training image
img_train = imread("img_train.png")
# remove unnecessary transparency column
img_train = img_train[:, :, :-1]
# convert to lab
img_train = rgb2hsv(img_train)
# normalize colors
#img_train = np.array(img_train, dtype=np.float64) / 255
# reshape image as list
w, h, d = tuple(img_train.shape)
train_array = np.reshape(img_train, (w * h, d))
# create train target for fitting the MLP classifier
train_target = np.empty((train_array.shape[0], ), dtype=int)
color_list = []
for i in range(len(train_target)):
if find_in_color_list(train_array[i], color_list) == -1:
color_list.append(train_array[i])
train_target[i] = find_in_color_list(train_array[i], color_list)