def test_write_load_json_file(self):
# prepare toy samples
self.haar_list.append(
haar(featureType.TWO_VERTICAL, (0, 0), 19, 19, 100000, 1))
self.haar_list.append(
haar(featureType.THREE_VERTICAL, (0, 0), 19, 19, 100000, 1, 1.5))
self.haar_list.append(
haar(featureType.FOUR, (0, 0), 19, 19, 100000, 1, .05))
# write toy samples to a json file
write_json_file(self.haar_list)
# load toy sample from a json file
self.classifiers = load_json_file()
for c in self.haar_list:
print(
"type: %s\tfeature type: %s\tposition: %s\twidth: %d\theight: %d\tthreshold: %d\tparity: %d\tweight: %f"
% (type(c), c.type.name, str(c.top_left), c.width, c.height,
c.threshold, c.parity, c.weight))
for c in self.classifiers:
print(
"type: %s\tfeature type: %s\tposition: %s\twidth: %d\theight: %d\tthreshold: %d\tparity: %d\tweight: %f"
% (type(c), c.type, str(c.top_left), c.width, c.height,
c.threshold, c.parity, c.weight))
assert len(self.haar_list) == len(self.classifiers)
for i in range(len(self.haar_list)):
assert two_haar_equal(self.haar_list[i], self.classifiers[i])
def _create_features(img_width, img_height, min_feat_width, max_feat_wid, min_feat_height, max_feat_height):
# private function to create all possible features
# return haar_feats: a list of haar-like features
# return type: np.array(haar.HaarLikeFeature)
haar_feats = list()
# iterate according to types of rectangle features
for feat_type in featureType:
# min of feature width is set in featureType enum
feat_start_width = max(min_feat_width, feat_type.value[0])
# iterate with step
for feat_width in range(feat_start_width, max_feat_wid, feat_type.value[0]):
# min of feature height is set in featureType enum
feat_start_height = max(min_feat_height, feat_type.value[1])
# iterate with setp
for feat_height in range(feat_start_height, max_feat_height, feat_type.value[1]):
# scan the whole image with sliding windows (both vertical & horizontal)
for i in range(img_width - feat_width):
for j in range(img_height - feat_height):
haar_feats.append(haar(feat_type, (i,j), feat_width, feat_height))
# haar_feats.append(haar(feat_type, (i,j), feat_width, feat_height, 0, -1)) # threshold = 0 (no misclassified images)
return haar_feats
def test_two_horizontal(self):
# check the Two-rectangle(horizontal) features
# included fail test in which parity == -1, indicating misclassification
feature = haar(featureType.TWO_HORIZONTAL, (0, 0), 24, 24, 100000, 1)
white = get_sum(self.int_img, (0, 0), (24, 12))
grey = get_sum(self.int_img, (0, 12), (24, 24))
expected = 1 if feature.threshold * feature.parity > white - grey else 0
expected_fail = 1 if feature.threshold * -1 > white - grey else 0
assert feature.get_vote(self.int_img) == expected
assert feature.get_vote(self.int_img) != expected_fail
def test_three_vertical(self):
# check the Three-rectangle(vertical) features
# included fail test in which parity == -1, indicating misclassification
feature = haar(featureType.THREE_VERTICAL, (0, 0), 24, 24, 100000, 1)
white = get_sum(self.int_img, (0, 0), (8, 24))
grey = get_sum(self.int_img, (0, 8), (24, 16))
white += get_sum(self.int_img, (0, 16), (24, 24))
expected = 1 if feature.threshold * feature.parity > white - grey else 0
expected_fail = 1 if feature.threshold * -1 > white - grey else 0
assert feature.get_vote(self.int_img) == expected
assert feature.get_vote(self.int_img) != expected_fail
def test_four(self):
# check the Four-rectangle features
# included fail test in which parity == -1, indicating misclassification
feature = haar(featureType.FOUR, (0, 0), 24, 24, 100000, 1)
white = get_sum(self.int_img, (0, 0), (12, 12))
grey = get_sum(self.int_img, (12, 0), (24, 12))
grey += get_sum(self.int_img, (0, 12), (12, 24))
white += get_sum(self.int_img, (12, 12), (24, 24))
expected = 1 if feature.threshold * feature.parity > white - grey else 0
expected_fail = 1 if feature.threshold * -1 > white - grey else 0
assert feature.get_vote(self.int_img) == expected
assert feature.get_vote(self.int_img) != expected_fail
def load_json_file():
# load the haar-like classifiers from a json file
# return: list of classifiers/features
# rtype: list[haar.HaarLikeFeature]
classifiers = list()
with open(path_json_file, 'r') as fp:
data_list = json.load(fp)
for i in range(len(data_list)):
data = data_list[str(i)]
classifiers.append(
haar(data["feature type"], data["position"], data["width"],
data["height"], data["threshold"], data["parity"],
data["weight"]))
return classifiers
def load_features(filename):
# Read data
read_features = list()
with open(path_save_features + filename, newline='') as features:
features_reader = csv.DictReader(features, delimiter='\t')
for feature in features_reader:
read_features.append(feature)
created_features = [
haar(featureType(eval(f['type'])),
eval(f['position']), eval(f['width']), eval(f['height']),
eval(f['error']), eval(f['threshold']), eval(f['polarity']),
eval(f['weight'])) for f in read_features
]
return created_features