def parameter_tuning_f1(dataset_path):
image_list = prefetch_images(dataset_path)
if len(image_list) <= 0:
return
param1 = (3, 5, 7, 9)
param2 = (16, 32, 64, 128, 256)
output = {}
num_samples = 40
for p1 in param1:
for p2 in param2:
print("*****" * 2, p1, p2, "*****" * 2)
features = {} # dictionary<family name, list of feature vector>
for i in range(len(image_list)):
family = image_list[i][0]
images = image_list[i][1]
# print("fetching...", family)
if len(images) < num_samples:
print("Insufficient samples")
return
for j in range(num_samples):
image_path = dataset_path + family + '/' + images[j]
# print(image_path, j + 1, "out of", num_samples)
img = io.imread(image_path)
# Feature 1: horizontal edge
hist = project_h_edge(img,
gauss_sigma=p1,
print_img=False,
total_blocks=p2)
# resize to desired length. e.g. 256
hist_arr = np.array(hist).reshape(1, len(hist))
feature_vec = transform.resize(hist_arr, (1, p2),
mode='reflect').tolist()[0]
if family not in features:
features[family] = list()
l = features.get(family)
l.append(feature_vec)
best_accuracy = fitting_scoring(features)
output[(p1, p2)] = best_accuracy
output = dict(
OrderedDict(
sorted(output.items(), key=lambda t: t[1], reverse=True)))
print(output)
best_key = list(output.keys())[0]
best_val = output[best_key]
print("-" * 20)
print("Feature1: param set", best_key, "accuracy", best_val)
print("-" * 20)
def parameter_tuning_f2(dataset_path):
image_list = prefetch_images(dataset_path)
if len(image_list) <= 0:
return
param1 = ((4, 4), (6, 6), (8, 8), (10, 10), (12, 12))
param2 = (16, 32, 64, 128, 256)
output = {}
num_samples = 40
for p1 in param1:
for p2 in param2:
print("*****" * 2, p1, p2, "*****" * 2)
features = {} # dictionary<family name, list of feature vector>
for i in range(len(image_list)):
family = image_list[i][0]
images = image_list[i][1]
print("fetching...", family)
if len(images) < num_samples:
print("Insufficient samples")
return
for j in range(num_samples):
image_path = dataset_path + family + '/' + images[j]
# print(image_path, j + 1, "out of", num_samples)
img = io.imread(image_path)
# Feature 2: Histogram of Gaussian
feature_vec = extract_HOG(img, blocks=p1,
buckets=p2).tolist()
if family not in features:
features[family] = list()
l = features.get(family)
l.append(feature_vec)
best_accuracy = fitting_scoring(features)
output[(p1, p2)] = best_accuracy
output = dict(
OrderedDict(
sorted(output.items(), key=lambda t: t[1], reverse=True)))
print(output)
best_key = list(output.keys())[0]
best_val = output[best_key]
print("-" * 20)
print("Feature 2: param set", best_key, "accuracy", best_val)
print("-" * 20)
def parameter_tuning_f3(dataset_path):
image_list = prefetch_images(dataset_path)
if len(image_list) <= 0:
return
param1 = ((1, 1), (2, 2), (4, 4), (8, 8), (16, 16))
output = {}
num_samples = 40
for p1 in param1:
print("*****" * 2, p1, "*****" * 2)
features = {} # dictionary<family name, list of feature vector>
for i in range(len(image_list)):
family = image_list[i][0]
images = image_list[i][1]
print("fetching...", family)
if len(images) < num_samples:
print("Insufficient samples")
return
for j in range(num_samples):
image_path = dataset_path + family + '/' + images[j]
# print(image_path, j + 1, "out of", num_samples)
img = io.imread(image_path)
# Feature 3: Mean intensity of each grid
grids = extract_grid_blocks(img, blocks_per_image=p1)
feature_vec = means_feature(grids)
if family not in features:
features[family] = list()
l = features.get(family)
l.append(feature_vec)
best_accuracy = fitting_scoring(features)
output[p1] = best_accuracy
output = dict(
OrderedDict(
sorted(output.items(), key=lambda t: t[1], reverse=True)))
print(output)
best_key = list(output.keys())[0]
best_val = output[best_key]
print("-" * 20)
print("Feature 3: param set", best_key, "accuracy", best_val)
print("-" * 20)
def parameter_tuning_f6(dataset_path):
image_list = prefetch_images(dataset_path)
if len(image_list) <= 0:
return
param1 = ((4, 4), (8, 8), (16, 16), (32, 32), (64, 64), (128, 128))
output = {}
num_samples = 40
for p1 in param1:
print("*****" * 2, p1, "*****" * 2)
features = {} # dictionary<family name, list of feature vector>
for i in range(len(image_list)):
family = image_list[i][0]
images = image_list[i][1]
print("fetching...", family)
if len(images) < num_samples:
print("Insufficient samples")
return
for j in range(num_samples):
image_path = dataset_path + family + '/' + images[j]
# print(image_path, j + 1, "out of", num_samples)
img = io.imread(image_path)
# Feature 6: Media images (flatten)
feature_vec = median_feature(img, resize_shape=p1)
if family not in features:
features[family] = list()
l = features.get(family)
l.append(feature_vec)
best_accuracy = fitting_scoring(features)
output[p1] = best_accuracy
output = dict(
OrderedDict(
sorted(output.items(), key=lambda t: t[1], reverse=True)))
print(output)
best_key = list(output.keys())[0]
best_val = output[best_key]
print("-" * 20)
print("Feature 6: param set", best_key, "accuracy", best_val)
print("-" * 20)
def parameter_tuning_f5(dataset_path):
image_list = prefetch_images(dataset_path)
if len(image_list) <= 0:
return
output = {}
num_samples = 40
print("*****" * 2, "*****" * 2)
features = {} # dictionary<family name, list of feature vector>
for i in range(len(image_list)):
family = image_list[i][0]
images = image_list[i][1]
print("fetching...", family)
if len(images) < num_samples:
print("Insufficient samples")
return
for j in range(num_samples):
image_path = dataset_path + family + '/' + images[j]
# print(image_path, j + 1, "out of", num_samples)
img = io.imread(image_path)
# Feature 5: Histogram(Contrast)
feature_vec = histogram_feature(img)
if family not in features:
features[family] = list()
l = features.get(family)
l.append(feature_vec)
best_accuracy = fitting_scoring(features)
output[256] = best_accuracy
output = dict(
OrderedDict(sorted(output.items(), key=lambda t: t[1], reverse=True)))
print(output)
best_key = list(output.keys())[0]
best_val = output[best_key]
print("-" * 20)
print("Feature 5: param set", best_key, "accuracy", best_val)
print("-" * 20)