def image_detach(color_img: np.ndarray, bin_num: int) -> np.ndarray:
"""
컬러 이미지 `color_img`를 색상에 따른 각 인스턴스 객체로 분리합니다.
Parameters
----------
color_img : np.ndarray
컬러 이미지
bin_num : int
분리할 최대 인스턴스 객체의 개수
Returns
-------
np.ndarray
`color_img`의 width, height에 `bin_num` channel인 `np.ndarray`
Examples
--------
>>> import cv2
>>> from utils.image_transform import image_detach
>>> img = cv2.imread("../simple_image_generator/samples/label.png")
>>> image_detach(img, 30)
"""
color_list: List[Tuple[int, int, int]] = list(
map(lambda v: v[0], get_all_colors(color_img, True)))
id_color_list: List[Tuple[int,
Tuple[int, int,
int]]] = color_map_generate(color_list)
return image_detach_with_id_color_list(color_img, id_color_list, bin_num)
def test_image_detach_with_id_color_list(self):
"""
`image_detach_with_id_color_list()` 메서드를 테스트합니다.
"""
# Prerequisite
image: np.ndarray = self.__get_sample_image()
image_colors: List[Tuple[int, int, int]] = list(
map(lambda v: v[0], image_info.get_all_colors(image, True))
)
image_color_map: List[
Tuple[int, Tuple[int, int, int]]
] = image_color_transform.color_map_generate(image_colors, True)
# Processing
bin_num: int = 30
mask_value: float = 1.0
detached_image: np.ndarray = (
image_color_transform.image_detach_with_id_color_list(
color_img=image,
id_color_list=image_color_map[:bin_num],
bin_num=bin_num,
mask_value=mask_value,
)
)
# Result
test_result_non_zero_counts: int = 32400
test_result_unique_values: Tuple[float, float] = [0.0, mask_value]
# Check
self.assertEqual(np.count_nonzero(detached_image), test_result_non_zero_counts)
self.assertEqual(np.unique(detached_image).tolist(), test_result_unique_values)
def test_color_map_generate_without_black(self):
"""
`color_map_generate()` 메서드를 테스트합니다.
검은 색을 추가하지 않는 옵션을 테스트합니다.
"""
# Prerequisite
image: np.ndarray = self.__get_lenna_image()
image_colors: List[Tuple[int, int, int]] = list(
map(lambda v: v[0], image_info.get_all_colors(image, True))
)
# Processing
image_color_map: List[
Tuple[int, Tuple[int, int, int]]
] = image_color_transform.color_map_generate(image_colors, False)
# Result
test_result_not_lenna_first_color_map: Tuple[Tuple[int, int, int]] = (
0,
(0, 0, 0),
)
# Check
self.assertNotEqual(
image_color_map[0][1], test_result_not_lenna_first_color_map
)
def test_get_levels_from_grayscale_image(self):
test_file_name: str = "bone.png"
test_image_fullpath: str = os.path.join(
self.test_path, self.test_resource_folder_name, test_file_name
)
gray_image = cv2.imread(test_image_fullpath, cv2.IMREAD_GRAYSCALE)
colors_on_image = image_info.get_all_colors(gray_image, True)
self.assertEqual(colors_on_image[0][0], (0,))
def test_get_all_colors__black_color_first(self):
test_file_name: str = "bone.png"
test_image_fullpath: str = os.path.join(
self.test_path, self.test_resource_folder_name, test_file_name
)
image = cv2.imread(test_image_fullpath)
colors_on_image: List[
Tuple[Tuple[int, int, int], int]
] = image_info.get_all_colors(image, True)
self.assertEqual(colors_on_image[0][0], (0, 0, 0))
def test_get_all_colors(self):
test_file_name: str = "lenna.png"
lenna_colors: int = 148279
test_image_fullpath: str = os.path.join(
self.test_path, self.test_resource_folder_name, test_file_name
)
image = cv2.imread(test_image_fullpath)
colors_on_image: List[
Tuple[Tuple[int, int, int], int]
] = image_info.get_all_colors(image, True)
self.assertEqual(len(colors_on_image), lenna_colors)
def test_image_detach_with_id_color_probability_list(self):
"""
`image_detach_with_id_color_probability_list()` 메서드를 테스트합니다.
"""
# Prerequisite
image: np.ndarray = self.__get_sample_image()
image_colors: List[Tuple[int, int, int]] = list(
map(lambda v: v[0], image_info.get_all_colors(image, True))
)
image_color_map: List[
Tuple[int, Tuple[int, int, int]]
] = image_color_transform.color_map_generate(image_colors, True)
# Processing
bin_num: int = 30
detached_image: np.ndarray = (
image_color_transform.image_detach_with_id_color_probability_list(
color_img=image,
id_color_list=image_color_map[:bin_num],
bin_num=bin_num,
resize_by_power_of_two=2,
)
)
# Result
test_result_non_zero_counts: int = 2189
test_result_unique_values: List[Tuple[float, int]] = [
(0.0, 58561),
(0.0625, 8),
(0.25, 152),
(0.4375, 8),
(0.5, 76),
(1.0, 1945),
]
# Check
self.assertEqual(np.count_nonzero(detached_image), test_result_non_zero_counts)
self.assertEqual(
np.unique(detached_image).tolist(),
list(map(lambda v: v[0], test_result_unique_values)),
)
for value in test_result_unique_values:
self.assertEqual(
np.count_nonzero(np.equal(detached_image, value[0])), value[1]
)
def test_color_map_generate_for_add_black(self):
"""
`color_map_generate()` 메서드를 테스트합니다.
검은 색이 없는 경우, 검은 색을 맨 앞으로 추가하는 옵션을 테스트합니다.
"""
# Prerequisite
image: np.ndarray = self.__get_lenna_image()
image_colors: List[Tuple[int, int, int]] = list(
map(lambda v: v[0], image_info.get_all_colors(image, True))
)
# Processing
image_color_map: List[
Tuple[int, Tuple[int, int, int]]
] = image_color_transform.color_map_generate(image_colors, True)
# Result
test_result_lenna_color_num: int = 148280
test_result_lenna_first_color_map: Tuple[Tuple[int, int, int]] = (0, (0, 0, 0))
# Check
self.assertEqual(len(image_color_map), test_result_lenna_color_num)
self.assertEqual(image_color_map[0], test_result_lenna_first_color_map)