def calc_dataset_size(train_path, test_path):
train_filenames = get_filenames(train_path, 'jpg')
test_filenames = get_filenames(test_path, 'jpg')
random.shuffle(train_filenames)
random.shuffle(test_filenames)
dataset_size = len(train_filenames)
return train_filenames, test_filenames, dataset_size
def normalize_imgs(imgs_arr, usr_imgs_path):
usr_img_paths = get_filenames(usr_imgs_path, 'dcm')
num_img_ls = []
for usr_img_path in usr_img_paths:
num_img = int(usr_img_path.split('/')[-1][:-4])
num_img_ls.append(num_img)
num_img_ls.sort()
# print(num_img_ls)
dist = len(num_img_ls) / imgs_arr.shape[0]
dataset_img_paths = []
for i in range(imgs_arr.shape[0]):
num_img_path = '{}/{}.dcm'.format(usr_imgs_path,
num_img_ls[int(i * dist)])
dataset_img_paths.append(num_img_path)
print(dataset_img_paths)
for i, dataset_img_path in enumerate(dataset_img_paths):
ct_dcm = pydicom.dcmread(dataset_img_path)
# print(ct_dcm.pixel_array)
num_mag = int(ct_dcm.pixel_array.shape[1] / imgs_arr.shape[2])
ct_img = ct_dcm.pixel_array[::num_mag, ::num_mag] / 65535
imgs_arr[i] = ct_img
return imgs_arr
def get_dataset_paths(usr_imgs_path, NUM_DIVIDED=20):
usr_img_paths = get_filenames(usr_imgs_path, 'dcm')
num_img_list = []
for usr_img_path in usr_img_paths:
num_img = int(usr_img_path.split('/')[-1][:-4])
num_img_list.append(num_img)
num_img_list.sort()
# print(num_img_list)
dist = len(num_img_list) / NUM_DIVIDED
dataset_img_paths = []
for i in range(NUM_DIVIDED):
num_img_path = '{}/{}.dcm'.format(usr_imgs_path,
num_img_list[int(i * dist)])
dataset_img_paths.append(num_img_path)
return dataset_img_paths
def normalize_imgs(imgs_arr, user_imgs_path):
user_img_paths = get_filenames(user_imgs_path, 'dcm')
num_img_list = []
for user_img_path in user_img_paths:
num_img = int(user_img_path.split('/')[-1][:-4])
num_img_list.append(num_img)
num_img_list.sort()
# print(num_img_list)
dist = len(num_img_list) / imgs_arr.shape[0]
dataset_img_paths = []
for i in range(imgs_arr.shape[0]):
num_img_path = '{}/{}.dcm'.format(user_imgs_path,
num_img_list[int(i * dist)])
dataset_img_paths.append(num_img_path)
# print(dataset_img_paths)
for i, dataset_img_path in enumerate(dataset_img_paths):
ct_dcm = pydicom.dcmread(dataset_img_path)
ct_img = transform_ctdata(ct_dcm, windowWidth=-1500, windowCenter=-600)
# print(ct_dcm.pixel_array)
j = i
if np.average(ct_img) < 25 or np.average(ct_img) > 200 or (
np.average(ct_img) == ct_img[:, :]).all():
print('pass')
j += 1
ct_dcm = pydicom.dcmread(dataset_img_paths[j])
ct_img = transform_ctdata(ct_dcm, -1500, -600)
lung_img = get_lung_img(ct_img.copy(), isShow=False)
output_img = get_square_img(lung_img)
output_img = cv2.resize(output_img,
(imgs_arr.shape[1], imgs_arr.shape[2]))
imgs_arr[i] = output_img
return imgs_arr
return
def cv2_winClose():
'''
Press "q" to close all cv2 windows
'''
cv2.waitKey(0)
if cv2.waitKey(1) & 0xFF == ord('q'): # 按q键退出
return
if __name__ == "__main__":
path = "Data/train_images/hog_data"
file_extension = "jpg"
filenames = get_filenames(path, file_extension)
steel = Hog_Param()
# positive data
for i in range(len(filenames)):
img = cv2.imread(filenames[i])
cv2.imshow('a', img)
cv2_winClose()
img = cv2.resize((128, 64))
hist = steel.hog.compute(img, (8, 8))
for j in range(steel.featureNum):
steel.featureArray[i, j] = hist[j]
steel.labelArray[i, 0] = 1