def exp():
image_dir = r"../images/medical_image_sets/DRIVE/images"
mask_dir = r"../images/medical_image_sets/DRIVE/mask"
out_dir = r"./temp_out/"
[os.remove(path) for path in glob.glob(out_dir + "/*")]
image_names = glob.glob(image_dir + "/*")
gamma_list = []
image_path_list = []
for k, path in enumerate(image_names):
# Step 1. read images
mask_path = os.path.join(mask_dir, path.split(os.sep)[-1])
img = cv2.imread(path)
mask = cv2.imread(mask_path, 0)
if img is None or img.shape[
2] == 4: # something wrong to read an image, or BGRA image, or other files
print("Warning: path (" + path +
") is not valid, we will skip this path...")
continue
_, mask = cv2.threshold(mask, 254, 255, cv2.THRESH_BINARY)
# Step 2. conduct gamma estimation and image restoration with different methods or config
_, img_cab = correct_average_brightness(img, mask=mask) # CAB
_, img_cab_agt = adaptive_gamma_transform(img, mask=mask,
visual=False) # AGT
gamma, img_cab_agt_visual = adaptive_gamma_transform(
img, mask=mask, visual=True) # AGT-ME-VISUAL
_, img_bigc = blind_inverse_gamma_correction(img, "fast", True) # BIGC
# Step 3. save the images
width = img.shape[0] # 196 for paper eps file
# names = ["ORIGINAL", "BIGC", "CAB", "AGT-ME", "AGT-ME-VISUAL"]
names = ["ORIGINAL", "BIGC", "CAB", "AGT-ME"]
images = [img, img_bigc, img_cab, img_cab_agt, img_cab_agt_visual]
all_imgs = None
for name, image in zip(names, images):
temp_img = cv2.resize(image, (0, 0),
fx=width / image.shape[1],
fy=width / image.shape[1])
# temp_img = cv2.putText(temp_img, name, (0, 25), cv2.FONT_HERSHEY_COMPLEX, 0.7, (255, 0, 0), 1)
temp_img = cv2.putText(temp_img, name, (0, 50),
cv2.FONT_HERSHEY_DUPLEX, 2.0, (255, 0, 0),
2)
if all_imgs is None:
all_imgs = temp_img.copy()
else:
all_imgs = np.concatenate(
[all_imgs, 255 + np.zeros([all_imgs.shape[0], 11, 3])],
axis=1)
all_imgs = np.concatenate([all_imgs, temp_img], axis=1)
all_imgs = all_imgs.astype(np.uint8)
cv2.imwrite(
os.path.join(out_dir,
path.split(os.sep)[-1][:-4] + ".png"), all_imgs)
os.system("nautilus " + out_dir)
def exp():
image_dir = r"../images/medical_image_sets/CT"
out_dir = r"./temp_out"
[os.remove(path) for path in glob.glob(out_dir + "/*")]
image_names = glob.glob(image_dir + "/*")
for k, img_path in enumerate(image_names):
print("processing " + img_path + "...")
# Step 1. read CT images with pydicom package
dcm = pydicom.read_file(img_path)
dcm.image = dcm.pixel_array * dcm.RescaleSlope + dcm.RescaleIntercept
img_raw = dcm.image.copy()
if img_raw is None: # something wrong to read an image, or BGRA image
print("Warning: path (" + img_path + ") is not valid, we will skip this path...")
continue
# Step 2. Get the mask with a simple threshold strategy and normalize it to 0-1
mask_threshold = -150
mask = ((img_raw > mask_threshold) * 255).astype(np.uint8)
img = 255 * (img_raw - mask_threshold) / (np.max(img_raw[:]) - mask_threshold)
img[img < 0] = 0
img = img.round().astype(np.uint8)
# Step 2. conduct gamma estimation and image restoration with different methods or config
_, img_cab = correct_average_brightness(img, mask=mask) # CAB
_, img_cab_agt = adaptive_gamma_transform(img, mask=mask, visual=False) # AGT
gamma, img_cab_agt_visual = adaptive_gamma_transform(img, mask=mask, visual=True) # AGT-ME-VISUAL
_, img_bigc = blind_inverse_gamma_correction(img, "fast", True) # BIGC
img_bigc[mask < 255] = 0
# Step 3. save the images
width = img.shape[0] # 196 for paper eps file
# names = ["ORIGINAL", "BIGC", "CAB", "AGT-ME", "AGT-ME-VISUAL"]
names = ["ORIGINAL", "BIGC", "CAB", "AGT-ME"]
images = [img, img_bigc, img_cab, img_cab_agt, img_cab_agt_visual]
all_imgs = None
for name, image in zip(names, images):
temp_img = cv2.resize(image, (0, 0), fx=width / image.shape[1], fy=width / image.shape[1])
temp_img = cv2.cvtColor(temp_img, cv2.COLOR_GRAY2BGR)
# temp_img = cv2.putText(temp_img, name, (0, 25), cv2.FONT_HERSHEY_COMPLEX, 0.7, (255, 0, 0), 1)
temp_img = cv2.putText(temp_img, name, (0, 50), cv2.FONT_HERSHEY_DUPLEX, 1.5, (255, 0, 0), 2)
if all_imgs is None:
all_imgs = temp_img.copy()
else:
all_imgs = np.concatenate([all_imgs, 255 + np.zeros([all_imgs.shape[0], 11, 3])], axis=1)
all_imgs = np.concatenate([all_imgs, temp_img], axis=1)
all_imgs = all_imgs.astype(np.uint8)
cv2.imwrite(os.path.join(out_dir, img_path.split(os.sep)[-1] + ".png"), all_imgs)
# cv2.imwrite(os.path.join(out_dir, img_path.split(os.sep)[-1] + "mask.png"), mask)
os.system("nautilus " + out_dir)
print("please check the results in dir:" + out_dir)
os.system("nautilus " + out_dir)
def exp():
image_dir = r"../images/medical_image_sets/MRI/"
out_dir = r"./temp_out"
[os.remove(path) for path in glob.glob(out_dir + "/*")]
image_names = glob.glob(image_dir + "/*")
for k, img_path in enumerate(image_names):
# Step 1. read images
img_raw = nib.load(img_path).dataobj[200:675, :, 5]
img_raw = cv2.rotate(img_raw, cv2.ROTATE_90_COUNTERCLOCKWISE)
mask = (img_raw > 100).astype(np.uint8) * 255
img = (255 * (img_raw - np.min(img_raw[:])) /
(np.max(img_raw[:]) - np.min(img_raw[:])))
img = img.round().astype(np.uint8)
# Step 2. conduct gamma estimation and image restoration with different methods or config
_, img_cab = correct_average_brightness(img, mask=mask) # CAB
_, img_cab_agt = adaptive_gamma_transform(img, mask=mask,
visual=False) # AGT
gamma, img_cab_agt_visual = adaptive_gamma_transform(
img, mask=mask, visual=True) # AGT-ME-VISUAL
_, img_bigc = blind_inverse_gamma_correction(img, "fast", True) # BIGC
img_bigc[mask < 255] = 0
# Step 3. save the images
width = img.shape[0] # 196 for paper eps file
# names = ["ORIGINAL", "BIGC", "CAB", "AGT-ME", "AGT-ME-VISUAL"]
names = ["ORIGINAL", "BIGC", "CAB", "AGT-ME"]
images = [img, img_bigc, img_cab, img_cab_agt, img_cab_agt_visual]
all_imgs = None
for name, image in zip(names, images):
temp_img = cv2.resize(image, (0, 0),
fx=width / image.shape[1],
fy=width / image.shape[1])
temp_img = cv2.cvtColor(temp_img, cv2.COLOR_GRAY2BGR)
# temp_img = cv2.putText(temp_img, name, (0, 25), cv2.FONT_HERSHEY_COMPLEX, 0.7, (255, 0, 0), 1)
temp_img = cv2.putText(temp_img, name, (0, 70),
cv2.FONT_HERSHEY_DUPLEX, 3.0, (255, 0, 0),
3)
if all_imgs is None:
all_imgs = temp_img.copy()
else:
all_imgs = np.concatenate(
[all_imgs, 255 + np.zeros([all_imgs.shape[0], 11, 3])],
axis=1)
all_imgs = np.concatenate([all_imgs, temp_img], axis=1)
all_imgs = all_imgs.astype(np.uint8)
cv2.imwrite(
os.path.join(out_dir,
img_path.split(os.sep)[-1][:-4] + ".png"), all_imgs)
os.system("nautilus " + out_dir)
def exp():
# Step.0 Set the image path
image_dir = r"../images/BSD68"
file_list = os.listdir(image_dir)[0:10]
size_list = [(256, 256), (512, 512), (1024, 1024), (2048, 2048)]
agt_time_list = []
cab_time_list = []
bigc_time_list = []
for k, name in enumerate(file_list):
print(str(k) + ":" + name)
image = cv2.imread(os.path.join(image_dir, name), cv2.IMREAD_GRAYSCALE)
if image is None:
continue
agt_time_list.append([])
cab_time_list.append([])
bigc_time_list.append([])
for siz in size_list:
img = cv2.resize(image, siz) # Resize the images
start_time = time.time()
_, _ = adaptive_gamma_transform(img)
end_time = time.time()
agt_time_list[-1].append(end_time - start_time)
start_time = time.time()
_, _ = correct_average_brightness(img)
end_time = time.time()
cab_time_list[-1].append(end_time - start_time)
start_time = time.time()
_, _ = blind_inverse_gamma_correction(img)
end_time = time.time()
bigc_time_list[-1].append(end_time - start_time)
agt_time_list = np.array(agt_time_list)
cab_time_list = np.array(cab_time_list)
bigc_time_list = np.array(bigc_time_list)
print("AGT time cost:")
print(" mean:" + str(np.mean(agt_time_list, axis=0)) + "(s)")
print(" std: " + str(np.std(agt_time_list, axis=0)) + "(s)")
print("CAB time cost:")
print(" mean:" + str(np.mean(cab_time_list, axis=0)) + "(s)")
print(" std: " + str(np.std(cab_time_list, axis=0)) + "(s)")
print("BIGC time cost:")
print(" mean:" + str(np.mean(bigc_time_list, axis=0)) + "(s)")
print(" std: " + str(np.std(bigc_time_list, axis=0)) + "(s)")
def exp():
# Step.0 Set the image dataset dir and a dir to store the results
image_dir = r"../images/BSD68/"
image_names = os.listdir(image_dir)[:MAX_NUMBER] # Get the file name of the test images
out_dir = r"./temp_out"
# Step.1 the gamma set definition
gamma_set = np.linspace(0.1, 3.0, 30)
for gamma in gamma_set: # save the distorted image for fun
path = os.path.join(image_dir, image_names[0])
img = cv2.imread(path, -1)
distorted_img = gamma_trans(img, gamma)
cv2.imwrite(out_dir + os.sep + str(gamma) + ".png", distorted_img)
# Step.2 get the estimated gamma with different methods
agt_gamma_estimated_list = []
cab_gamma_estimated_list = []
bigc_gamma_estimated_list = []
for gamma in gamma_set:
agt_gamma_estimated_list.append([])
cab_gamma_estimated_list.append([])
bigc_gamma_estimated_list.append([])
for k, name in enumerate(image_names):
print("gamma:" + str(gamma) + "; number:" + str(k) + ": " + name)
path = os.path.join(image_dir, name)
img = cv2.imread(path, cv2.IMREAD_GRAYSCALE) # read as gray image
# Test with AGT method
gamma_origin, _ = adaptive_gamma_transform(img, visual=False) # Estimate the original gamma value
distorted_img = gamma_trans(img, gamma) # Add gamma distortion
gamma_estimated, _ = adaptive_gamma_transform(distorted_img, visual=False) # distorted gamma value
agt_gamma_estimated_list[-1].append(gamma_origin / gamma_estimated)
# Test with CAB method
gamma_origin, _ = correct_average_brightness(img) # Estimate the original gamma value
distorted_img = gamma_trans(img, gamma) # Add gamma distortion
gamma_estimated, _ = correct_average_brightness(distorted_img) # distorted gamma value
cab_gamma_estimated_list[-1].append(gamma_origin / gamma_estimated)
# Test with BIGC method
gamma_origin, _ = blind_inverse_gamma_correction(img, visual=False) # Estimate the original gamma value
distorted_img = gamma_trans(img, gamma) # Add gamma distortion
gamma_estimated, _ = blind_inverse_gamma_correction(distorted_img, visual=False) # distorted gamma value
bigc_gamma_estimated_list[-1].append(gamma_origin / gamma_estimated)
agt_gamma_estimated_list = np.array(agt_gamma_estimated_list)
cab_gamma_estimated_list = np.array(cab_gamma_estimated_list)
bigc_gamma_estimated_list = np.array(bigc_gamma_estimated_list)
# figure 1. actual gamma VS estimated gamma (AGT method)
plt.figure()
plt.subplots_adjust(bottom=0.14, left=0.14)
plt.grid()
gamma_set = np.repeat(gamma_set[:, np.newaxis], agt_gamma_estimated_list.shape[1], axis=1)
plt.scatter(gamma_set, agt_gamma_estimated_list, s=20, c='b')
plt.xlim([-0.2, 3.2])
plt.ylim([-0.2, 3.2])
plt.xlabel("Gamma bias $\gamma_b$", fontsize=16)
plt.ylabel("Recognized gamma $\gamma_r$", fontsize=16)
plt.tick_params(labelsize=16)
# plt.legend(fontsize=20)
foo_fig = plt.gcf()
foo_fig.savefig("figs/Fig4a.eps", format='eps', dpi=1200)
plt.title("set-estimate")
# figure 2. RMSE error curve for different methods
rmse_agt = np.sqrt(np.mean((agt_gamma_estimated_list - gamma_set) ** 2, axis=1))
rmse_cab = np.sqrt(np.mean((cab_gamma_estimated_list - gamma_set) ** 2, axis=1))
temp = (bigc_gamma_estimated_list - gamma_set) ** 2
print("Outlier percentage of BIGC:", 100.0 * np.sum(temp[:] > 0.25) / temp.size, "%")
temp[temp > 0.5] = np.NaN # special treatment for BIGC outliers
rmse_bigc = np.sqrt(np.nanmean(temp, axis=1))
rmse_agt[rmse_agt < 1e-3] = 1e-3
rmse_cab[rmse_cab < 1e-3] = 1e-3
rmse_bigc[rmse_bigc < 1e-3] = 1e-3
print("mean rmse:", np.mean(rmse_agt[:]), "(AGT-ME) and ", np.mean(rmse_cab[:]), "(CAB) and ",
np.mean(rmse_bigc[:]), "(BIGC)")
# print(rmse_agt, rmse_bigc)
plt.figure()
plt.grid()
plt.subplots_adjust(bottom=0.14, left=0.18)
plt.plot(gamma_set[:, 0], rmse_bigc, "r-.", label="BIGC")
plt.plot(gamma_set[:, 0], rmse_cab, "r:", label="CAB")
plt.plot(gamma_set[:, 0], rmse_agt, "b-", label="AGT-ME")
plt.xlim([0.0, 3.0])
plt.ylim([1e-3, 1.0])
plt.semilogy()
plt.xlabel("Gamma bias $\gamma_b$", fontsize=16)
plt.ylabel("RMSE", fontsize=16)
plt.tick_params(labelsize=16)
plt.legend(fontsize=16)
foo_fig = plt.gcf()
foo_fig.savefig("figs/Fig4b.eps", format='eps', dpi=1200)
plt.show()
def line_value(image):
return image[line_index, 250:850].astype(np.float)
# Step.0 path setting
image_dir = r"../images/slm image"
out_dir = r"./temp_out"
[os.remove(path) for path in glob.glob(out_dir + "/*")]
# Step.1 read images
path = glob.glob(image_dir + "/*")[0]
img = cv2.imread(path, cv2.IMREAD_GRAYSCALE) # read as gray image
# Step 2. conduct gamma estimation and image restoration with different methods or config
_, img_cab = correct_average_brightness(img) # CAB
_, img_agt = adaptive_gamma_transform(img, visual=False) # AGT
gamma, img_agt_visual = adaptive_gamma_transform(img,
visual=True) # AGT-ME-VISUAL
_, img_bigc = blind_inverse_gamma_correction(img, "fast", True) # BIGC
# Step.2 power spectrum analysis
names = ["ORIGINAL", "BIGC", "AGT-ME", "CAB"]
images = [img, img_bigc, img_agt, img_cab]
styles = ["g-", "r-.", "b-", "r:"]
line_index = 50 # 200
power_spec = [spectrum(temp) for temp in images]
line_values = [line_value(temp) for temp in images]
# Step.3 display the curves (intensity curve and power spectrum)
plt.figure(figsize=(6, 3))