def exudates_get_features_3(folder, elements_folder, fundus_mask, num_imgs=10,
stride_width=6):
thr_high = 189.0 / 255.0
y = []
pxl_df = pd.DataFrame(columns=['hue', 'intensity', 'mean intensity', 'std intensity', 'dist to optic disk'])
for img_fn in image_names_in_folder(folder)[:num_imgs]:
print 'Image filename:', img_fn
img = load_image(img_fn)
short_fn = os.path.split(img_fn)[-1]
hard_exud = load_image(os.path.join(elements_folder, short_fn))
hsi = rgb_to_hsi(img)
intensity = hsi[:, :, 2].copy()
#i_sp = add_salt_and_pepper(intensity, 0.005)
i_med = mh.median_filter(intensity) # use Wiener filter instead?
i_clahe = skimage.exposure.equalize_adapthist(i_med)
optic_disk_map = optic_disk_detect_3(img)
mask, optic_disk_center = optic_disk_mask(optic_disk_map, return_center=True)
print 'Disk center: ', optic_disk_center
fundus_wo_disk_mask = fundus_mask & (~mask)
pxl_mask = np.zeros_like(i_clahe, dtype=np.bool)
pxl_mask[::stride_width, ::stride_width] = True
pxl_mask[~fundus_wo_disk_mask] = False
bbox = bounding_box(hard_exud >= thr_high, bbox_radius_ratio=2)
pxl_mask &= bbox
cur_pxl_df = pd.DataFrame(columns=['hue', 'intensity', 'mean intensity', 'std intensity', 'dist to optic disk'])
cur_pxl_df['hue'] = hsi[:, :, 0][pxl_mask].ravel()
cur_pxl_df['intensity'] = i_clahe[pxl_mask].ravel()
cur_pxl_df['mean intensity'] = pf.pxls_mean_intensity(i_clahe, stride_width, neighbd_rad=7, mask=pxl_mask).ravel()
cur_pxl_df['std intensity'] = pf.pxls_std_intensity(i_clahe, stride_width, neighbd_rad=3, mask=pxl_mask).ravel()
cur_pxl_df['dist to optic disk'] = pf.pxls_dist_to_pnt(i_clahe.shape, optic_disk_center,
stride_width, mask=pxl_mask).ravel()
print 'Accounted pixels: ', len(cur_pxl_df)
y.append(hard_exud[pxl_mask]) # we can use fraction of white points in a neighborhood instead,
# but it becomes regression task then
pxl_df = pd.concat((pxl_df, cur_pxl_df), axis=0, ignore_index=True)
y = np.hstack(y)
y = y >= thr_high
return pxl_df, y
def get_clusters_from_img_set(folder, box_shape=30):
boxes = []
for img_fn in sorted(glob.glob(os.path.join(folder, '*.jpeg'))):
img = load_image(img_fn)
gb = prepare_img(img)
t = np.percentile(gb.ravel(), 99)
bright = gb > t
bright = (bright * 255).astype(int)
cl1 = bfs_clustered(bright, radius=2, tol=1)
cl2 = erase_clusters(cl1,
small_tol=0.0004 * cl1.shape[0] * cl1.shape[1],
large_tol=2 * cl1.size,
small_replace=0)
cl_boxes, _ = get_clusters(cl2, box_shape=box_shape)
cl_boxes = cl_boxes[1:] # ignoring background
cl_boxes = cl_boxes[cl_boxes.mean(axis=1) != 0.0]
boxes.append(cl_boxes)
boxes = np.array(boxes)
boxes = np.vstack(boxes)
return boxes
def extract_RIM_ONE_v2(db_folder):
"""
Cropped (to optic disc region) images with polygonal optic disc segmentation.
380 x 394 original, 394 x 394 after post-processing.
Required schema:
db_folder/
Normal/
im{:03}.jpg (number from 0 to 255)
im{:03}_gs.txt
Glaucoma and glaucoma suspicious/
im{:03}.jpg (number from 256 to 455)
im{:03}_gs.txt
"""
orig_resolution = (380, 394)
result_resolution = (394, 394)
X_all, Y_all, filecodes_all, is_ill = [], [], [], []
for pic_type in ('Normal', 'Glaucoma and glaucoma suspicious'):
X, filenames = imh.load_set(os.path.join(db_folder, pic_type))
file_codes = [fn[-7:-4] for fn in filenames]
Y = []
for i, code in enumerate(file_codes):
anot_filename = os.path.join(db_folder, pic_type,
'Im{}-gs.txt'.format(code))
with open(anot_filename) as anot_fin:
lines = anot_fin.readlines()
'''
# polygonal segmentation
coords = lines[1:lines.index('Ellipse parameters\r\n')]
coords = np.array(map(int, coords))
if coords.size % 2 != 0:
raise imh.ImLibException('n_coords % 2 != 0')
coords = coords.reshape((coords.size / 2, 2))
coords = coords[1:] # optic disc center point is included in annotation for some reason
segm_img = np.zeros(orig_resolution, dtype=np.uint8)
cv2.fillPoly(segm_img, coords.reshape((1,) + coords.shape), color=1)
'''
'''
# ellipse segmentation
coords = lines[lines.index('Ellipse parameters\r\n') + 1:]
coords = map(int, coords)
i0, j0, a, b, angle = coords
a /= 2
b /= 2
segm_img = np.zeros(orig_resolution, dtype=np.uint8)
cv2.ellipse(segm_img, (i0, j0), (a, b), angle, 0, 360, color=1, thickness=-1)
'''
# acquiring segmentation from pre-computed image
segm_img = imh.load_image(
os.path.join(db_folder, pic_type + ' segmentation',
'Im{}-gs_mask.jpg'.format(code)))
Y.append(segm_img)
is_ill.append(HEALTHY if pic_type ==
'Normal' else GLAUCOMA_OR_SUSPECT)
for i in xrange(len(X)):
side = result_resolution[0]
X[i] = imh.resize_image_to_square(X[i], side, pad_cval=0)
Y[i] = imh.resize_image_to_square(Y[i], side, pad_cval=0)
Y[i] = Y[i].reshape(Y[i].shape + (1, ))
X_all.extend(X)
Y_all.extend(Y)
filecodes_all.extend(file_codes)
return X_all, Y_all, filecodes_all, is_ill
def extract_DRISHTI_GS_train(db_folder, return_disc=True, return_cup=True):
"""
Full images with optic disc and optic cup segmentation.
Average segmentation and "softmap" segmentation image are given.
50 images of various resolution close to 2040 x 1740.
Data set is split into training and test sets. Groundtruth is available for training set only.
This function returns Training set only.
Required schema:
db_folder/
Drishti-GS1_files/
Training/
Images/
drishtiGS_{:03}.png # some numbers are omitted, like 001, 003, 004, ...
GT/
drishtiGS_{:03}/
drishtiGS_{:03}_cdrValues.txt
AvgBoundary/
drishtiGS_{:03}_ODAvgBoundary.txt
drishtiGS_{:03}_CupAvgBoundary.txt
drishtiGS_{:03}_diskCenter.txt
SoftMap/
drishtiGS_{:03}_ODsegSoftmap.png
drishtiGS_{:03}_cupsegSoftmap.png
"""
result_resolution = (2040, 2040)
disc_all, cup_all, file_codes_all = [], [], []
set_path = os.path.join(db_folder, 'Drishti-GS1_files', 'Training')
images_path = os.path.join(set_path, 'Images')
X_all, file_names = imh.load_set(images_path)
rel_file_names = [os.path.split(fn)[-1] for fn in file_names]
rel_file_names_wo_ext = [fn[:fn.rfind('.')] for fn in rel_file_names]
file_codes = [
'Training' + fn[fn.find('_'):] for fn in rel_file_names_wo_ext
]
file_codes_all.extend(file_codes)
for fn in rel_file_names_wo_ext:
if return_disc:
disc_segmn = imh.load_image(
os.path.join(set_path, 'GT', fn, 'SoftMap',
fn + '_ODsegSoftmap.png'))
disc_all.append(disc_segmn)
if return_cup:
cup_segmn = imh.load_image(
os.path.join(set_path, 'GT', fn, 'SoftMap',
fn + '_cupsegSoftmap.png'))
cup_all.append(cup_segmn)
for i in xrange(len(X_all)):
side = result_resolution[0]
X_all[i] = imh.resize_image_to_square(X_all[i], side, pad_cval=0)
if return_disc:
disc_all[i] = imh.resize_image_to_square(disc_all[i],
side,
pad_cval=0)
disc_all[i] = disc_all[i].reshape(disc_all[i].shape + (1, ))
if return_cup:
cup_all[i] = imh.resize_image_to_square(cup_all[i],
side,
pad_cval=0)
cup_all[i] = cup_all[i].reshape(cup_all[i].shape + (1, ))
if return_disc:
if return_cup:
return X_all, disc_all, cup_all, file_codes_all
return X_all, disc_all, file_codes_all
if return_cup:
return X_all, cup_all, file_codes_all
return X_all, file_codes_all
def extract_RIM_ONE_v3(db_folder,
expert='avg',
return_disc=True,
return_cup=True):
"""
Cropped (to optic disc region, and a little more by vertical axis) images
with polygonal optic disc segmentation. 1424 x 2144 original, 1424 x 1424 after post-processing.
Images are two-channel (stereo) --- caught from 2 angles.
But segmentation is given for only one view (see L/R letter in file name for clarification).
So only one view of two is chosen.
Accepted values for `expert`: 1, 2, 'avg'.
Required schema:
db_folder/
Healthy/
Stereo Images/
N-{}-[L,R].jpg (number is without leading zeros, from 1 to 92)
(image cannot be used as is. it is two-part image, divided by vertical border)
Expert1_masks/
N-{}-[L,R]-Cup-exp1.png (4 files for one image number and L/R characteristic)
N-{}-[L,R]-Cup-exp1.txt
N-{}-[L,R]-Disc-exp1.png
N-{}-[L,R]-Disc-exp1.txt
Expert2_masks/
N-{}-[L,R]-Cup-exp2.png (4 files for one image number and L/R characteristic)
N-{}-[L,R]-Cup-exp2.txt
N-{}-[L,R]-Disc-exp2.png
N-{}-[L,R]-Disc-exp2.txt
Average_masks/
N-{}-[L,R]-Cup-Avg.png (4 files for one image number and L/R characteristic)
N-{}-[L,R]-Cup-Avg.txt
N-{}-[L,R]-Disc-Avg.png
N-{}-[L,R]-Disc-Avg.txt
Glaucoma and suspects/
(...) (the same as for Healthy, but images start with G not N)
"""
orig_resolution = (1424, 2144)
result_resolution = (1424, 1424)
if expert == 1:
expert_folder = 'Expert1_masks'
suffix = 'exp1'
elif expert == 2:
expert_folder = 'Expert2_masks'
suffix = 'exp2'
elif expert == 'avg':
expert_folder = 'Average_masks'
suffix = 'Avg'
else:
raise imh.ImLibException('value for "expert" argument not understood')
X_all, disc_all, cup_all, file_codes_all, is_ill = [], [], [], [], []
for pic_type in ('Healthy', 'Glaucoma and suspects'):
X, file_names = imh.load_set(
os.path.join(db_folder, pic_type, 'Stereo Images'))
X_all.extend(X)
rel_file_names = [os.path.split(fn)[-1] for fn in file_names]
file_codes = [fn[:fn.rfind('.')] for fn in rel_file_names]
file_codes_all.extend(file_codes)
for fc in file_codes:
if return_disc:
disc_segmn = imh.load_image(
os.path.join(db_folder, pic_type, expert_folder,
'{}-Disc-{}.png'.format(fc, suffix)))
disc_all.append(disc_segmn)
if return_cup:
cup_segmn = imh.load_image(
os.path.join(db_folder, pic_type, expert_folder,
'{}-Cup-{}.png'.format(fc, suffix)))
cup_all.append(cup_segmn)
is_ill.append(HEALTHY if pic_type ==
'Healthy' else GLAUCOMA_OR_SUSPECT)
for i in xrange(len(X_all)):
side = result_resolution[0]
if file_codes_all[i][-1] == 'L':
X_all[i] = X_all[i][:, :orig_resolution[1] / 2]
elif file_codes_all[i][-1] == 'R':
X_all[i] = X_all[i][:, orig_resolution[1] / 2:]
if return_disc:
disc_all[i] = disc_all[i][:, :orig_resolution[1] / 2]
if return_cup:
cup_all[i] = cup_all[i][:, :orig_resolution[1] / 2]
else:
raise imh.ImLibException(
'image {} has no L/R characteristic'.format(file_codes_all[i]))
X_all[i] = imh.resize_image_to_square(X_all[i], side, pad_cval=0)
if return_disc:
disc_all[i] = imh.resize_image_to_square(disc_all[i],
side,
pad_cval=0)
disc_all[i] = disc_all[i].reshape(disc_all[i].shape + (1, ))
if return_cup:
cup_all[i] = imh.resize_image_to_square(cup_all[i],
side,
pad_cval=0)
cup_all[i] = cup_all[i].reshape(cup_all[i].shape + (1, ))
if return_disc:
if return_cup:
return X_all, disc_all, cup_all, file_codes_all, is_ill
return X_all, disc_all, file_codes_all, is_ill
if return_cup:
return X_all, cup_all, file_codes_all, is_ill
return X_all, file_codes_all, is_ill