def compute_color2d_superpixels_features(image,
dict_features,
sp_size=30,
sp_regul=0.2):
""" segment image into superpixels and estimate features per superpixel
:param ndarray image: input RGB image
:param dict(list(str)) dict_features: list of features to be extracted
:param int sp_size: initial size of a superpixel(meaning edge length)
:param float sp_regul: regularisation in range(0;1) where "0" gives elastic
and "1" nearly square segments
:return list(list(int)), [[floats]]: superpixels and related of features
"""
assert sp_regul > 0., 'slic. regularisation must be positive'
logging.debug('run Superpixel clustering.')
slic = segment_slic_img2d(image,
sp_size=sp_size,
relative_compact=sp_regul)
# plt.figure(), plt.imshow(slic)
logging.debug('extract slic/superpixels features.')
features, _ = compute_selected_features_img2d(image, slic, dict_features)
logging.debug('list of features RAW: %r', features.shape)
features[np.isnan(features)] = 0
# if fts_norm:
# logging.debug('norm all features.')
# features, _ = seg_fts.norm_features(features)
# logging.debug('list of features NORM: %s', repr(features.shape))
return slic, features
def prepare_boundary_points_close(seg,
centers,
sp_size=25,
relative_compact=0.3):
""" extract some point around foreground boundaries
:param ndarray seg: input segmentation
:param [(int, int)] centers: list of centers
:param int sp_size: superpixel size
:return [ndarray]:
>>> seg = np.zeros((100, 200), dtype=int)
>>> ell_params = 50, 100, 40, 60, np.deg2rad(30)
>>> seg = add_overlap_ellipse(seg, ell_params, 1)
>>> pts = prepare_boundary_points_close(seg, [(40, 90)])
>>> pts # doctest: +NORMALIZE_WHITESPACE +ELLIPSIS
[array([[ 6, 85],
[ 8, 150],
...
[ 92, 118]])]
"""
slic = segment_slic_img2d(seg / float(seg.max()),
sp_size=sp_size,
relative_compact=relative_compact)
points_all = filter_boundary_points(seg, slic)
dists = spatial.distance.cdist(points_all, centers, metric='euclidean')
close_center = np.argmin(dists, axis=1)
points_centers = []
for i in range(int(close_center.max() + 1)):
points = points_all[close_center == i]
points_centers.append(points)
return points_centers
def prepare_boundary_points_close(seg, centers, sp_size=25, rltv_compact=0.3):
""" extract some point around foreground boundaries
:param ndarray seg: input segmentation
:param [(int, int)] centers: list of centers
:return [ndarray]:
>>> seg = np.zeros((100, 200), dtype=int)
>>> ell_params = 50, 100, 40, 60, np.deg2rad(30)
>>> seg = add_overlap_ellipse(seg, ell_params, 1)
>>> pts = prepare_boundary_points_close(seg, [(40, 90)])
>>> sorted(np.round(pts).tolist()) # doctest: +NORMALIZE_WHITESPACE
[[[6, 85], [8, 150], [16, 109], [27, 139], [32, 77], [36, 41], [34, 177],
[59, 161], [54, 135], [67, 62], [64, 33], [84, 150], [91, 48], [92, 118]]]
"""
slic = seg_spx.segment_slic_img2d(seg / float(seg.max()),
sp_size=sp_size,
rltv_compact=rltv_compact)
points_all = filter_boundary_points(seg, slic)
dists = spatial.distance.cdist(points_all, centers, metric='euclidean')
close_center = np.argmin(dists, axis=1)
points_centers = []
for i in range(int(close_center.max() + 1)):
points = points_all[close_center == i]
points_centers.append(points)
return points_centers
def compute_boundary_distance(idx_row, params, path_out=''):
""" compute nearest distance between two segmentation contours
:param (int, str) idx_row:
:param {} params:
:param str path_out:
:return (str, float):
"""
_, row = idx_row
name = os.path.splitext(os.path.basename(row['path_image']))[0]
img = load_image(row['path_image'], params['img_type'])
segm = load_image(row['path_segm'], 'segm')
logging.debug('segment SLIC...')
slic = seg_spx.segment_slic_img2d(img,
params['slic_size'],
params['slic_regul'],
params['slico'])
_, dists = seg_lbs.compute_boundary_distances(segm, slic)
if os.path.isdir(path_out):
logging.debug('visualise results...')
fig = tl_visu.figure_segm_boundary_dist(segm, slic)
fig.savefig(os.path.join(path_out, name + '.jpg'))
return name, np.mean(dists)
def load_inputs(name):
img, _ = load_image_2d(os.path.join(PATH_IMAGE, name + '.jpg'))
seg, _ = load_image_2d(os.path.join(PATH_SEGM, name + '.png'))
annot, _ = load_image_2d(os.path.join(PATH_ANNOT, name + '.png'))
centers = pd.read_csv(os.path.join(PATH_CENTRE, name + '.csv'), index_col=0).values
centers[:, [0, 1]] = centers[:, [1, 0]]
slic = segment_slic_img2d(img, sp_size=25, relative_compact=0.3)
return img, seg, slic, centers, annot
def load_image_annot_compute_features_labels(idx_row,
params,
show_debug_imgs=SHOW_DEBUG_IMAGES
):
""" load image and annotation, and compute superpixel features and labels
:param (int, {...}) idx_row: row from table with paths
:param {str: ...} params: segmentation parameters
:param bool show_debug_imgs: whether show debug images
:return (...):
"""
def _path_out_img(params, dir_name, name):
return os.path.join(params['path_exp'], dir_name, name + '.png')
idx, row = idx_row
idx_name = get_idx_name(idx, row['path_image'])
img = load_image(row['path_image'], params['img_type'])
annot = load_image(row['path_annot'], '2d_segm')
logging.debug('.. processing: %s', idx_name)
assert img.shape[:2] == annot.shape[:2], \
'individual size of image %s and seg_pipe %s for "%s" - "%s"' % \
(repr(img.shape), repr(annot.shape), row['path_image'], row['path_annot'])
if show_debug_imgs:
plt.imsave(_path_out_img(params, FOLDER_IMAGE, idx_name),
img,
cmap=plt.cm.gray)
plt.imsave(_path_out_img(params, FOLDER_ANNOT, idx_name), annot)
# duplicate gray band to be as rgb
# if img.ndim == 2:
# img = np.rollaxis(np.tile(img, (3, 1, 1)), 0, 3)
slic = seg_spx.segment_slic_img2d(img,
sp_size=params['slic_size'],
rltv_compact=params['slic_regul'])
img = tl_data.convert_img_color_from_rgb(img,
params.get('clr_space', 'rgb'))
logging.debug('computed SLIC with %i labels', slic.max())
if show_debug_imgs:
img_rgb = use_rgb_image(img)
img_slic = segmentation.mark_boundaries(img_rgb,
slic,
color=(1, 0, 0),
mode='subpixel')
plt.imsave(_path_out_img(params, FOLDER_SLIC, idx_name),
np.clip(img_slic, 0, 1))
slic_label_hist = seg_label.histogram_regions_labels_norm(slic, annot)
labels = np.argmax(slic_label_hist, axis=1)
slic_annot = labels[slic]
if show_debug_imgs:
plt.imsave(_path_out_img(params, FOLDER_SLIC_ANNOT, idx_name),
np.clip(slic_annot, 0, slic_annot.max()))
features, feature_names = seg_fts.compute_selected_features_img2d(
img, slic, params['features'])
return idx_name, img, annot, slic, features, labels, \
slic_label_hist, feature_names
def test_region_growing_graphcut(self, name='insitu7545'):
""" """
if not os.path.exists(PATH_PKL_MODEL):
self.test_shape_modeling()
# file_model = pickle.load(open(PATH_PKL_MODEL, 'r'))
npz_file = np.load(PATH_PKL_MODEL, allow_pickle=True)
file_model = dict(npz_file[npz_file.files[0]].tolist())
logging.info('loaded model: %r', file_model.keys())
list_mean_cdf = file_model['cdfs']
model = file_model['mix_model']
img, _ = load_image_2d(os.path.join(PATH_IMAGE, name + '.jpg'))
seg, _ = load_image_2d(os.path.join(PATH_SEGM, name + '.png'))
annot, _ = load_image_2d(os.path.join(PATH_ANNOT, name + '.png'))
centers = pd.read_csv(os.path.join(PATH_CENTRE, name + '.csv'), index_col=0).values
centers[:, [0, 1]] = centers[:, [1, 0]]
slic = segment_slic_img2d(img, sp_size=25, relative_compact=0.3)
slic_prob_fg = compute_segm_prob_fg(slic, seg, LABELS_FG_PROB)
dict_debug = {}
labels_gc = region_growing_shape_slic_graphcut(
slic,
slic_prob_fg,
centers, (model, list_mean_cdf),
'set_cdfs',
coef_shape=5.,
coef_pairwise=15.,
prob_label_trans=[0.1, 0.03],
optim_global=False,
nb_iter=65,
allow_obj_swap=False,
dict_thresholds=DEFAULT_RG2SP_THRESHOLDS,
debug_history=dict_debug
)
segm_obj = labels_gc[slic]
logging.info('show debug: %r', dict_debug.keys())
for i in np.linspace(0, len(dict_debug['criteria']) - 1, 5):
fig = figure_rg2sp_debug_complete(seg, slic, dict_debug, int(i), max_size=5)
fig_name = 'RG2Sp_graph-cut_%s_debug-%03d.pdf' % (name, i)
fig.savefig(os.path.join(PATH_OUTPUT, fig_name), bbox_inches='tight', pad_inches=0)
plt.close(fig)
_ = adjusted_rand_score(annot.ravel(), segm_obj.ravel())
# self.assertGreaterEqual(score, 0.5)
expert_segm(name, img, seg, segm_obj, annot, str_type='RG2Sp_graph-cut')
def test_general(self):
slic = segment_slic_img2d(self.img, sp_size=15, relative_compact=0.2)
logging.debug(np.max(slic))
vertices, edges = make_graph_segm_connect_grid2d_conn4(slic)
logging.debug('vertices: %r', vertices)
logging.debug(len(edges))
logging.debug('edges: %r', edges)
fig, axarr = plt.subplots(ncols=2)
axarr[0].imshow(self.img)
axarr[1].imshow(slic, cmap=plt.cm.jet)
fig.savefig(os.path.join(PATH_OUTPUT, 'temp_superpixels.png'))
if logging.getLogger().getEffectiveLevel() == logging.DEBUG:
plt.show()
plt.close(fig)
def get_slic_points_labels(segm, img=None, slic_size=20, slic_regul=0.1):
""" run SLIC on image or supepixels and return superpixels, their centers
and also lebels (label from segmentation in position of superpixel centre)
:param ndarray segm: segmentation
:param ndarray img: input image
:param int slic_size: superpixel size
:param float slic_regul: regularisation in range (0, 1)
:return tuple:
"""
if not img:
img = segm / float(segm.max())
slic = segment_slic_img2d(img, sp_size=slic_size, relative_compact=slic_regul)
slic_centers = np.array(superpixel_centers(slic)).astype(int)
labels = segm[slic_centers[:, 0], slic_centers[:, 1]]
return slic, slic_centers, labels
def test_show_image_features_clr2d(self):
img = load_sample_image(IMAGE_LENNA)
img = transform.resize(img, (128, 128))
slic = segment_slic_img2d(img, sp_size=10, relative_compact=0.2)
features, names = compute_selected_features_color2d(
img, slic, FEATURES_SET_ALL)
path_dir = os.path.join(PATH_OUTPUT, 'temp_image-rgb2d-features')
if not os.path.isdir(path_dir):
os.mkdir(path_dir)
for i in range(features.shape[1]):
fts = features[:, i]
im_fts = fts[slic]
p_fig = os.path.join(path_dir, names[i] + '.png')
plt.imsave(p_fig, im_fts)
self.assertTrue(os.path.isfile(p_fig))
def test_count_transitions_segment(self):
img = self.img[:, :, 0]
annot = self.annot.astype(int)
slic = segment_slic_img2d(img, sp_size=15, relative_compact=0.2)
label_hist = histogram_regions_labels_norm(slic, annot)
labels = np.argmax(label_hist, axis=1)
trans = count_label_transitions_connected_segments({'a': slic}, {'a': labels})
path_csv = os.path.join(PATH_OUTPUT, 'labels_transitions.csv')
pd.DataFrame(trans).to_csv(path_csv)
gc_regul = compute_pairwise_cost_from_transitions(trans, 10.)
np.random.seed(0)
features = np.tile(labels, (5, 1)).T.astype(float)
features += np.random.random(features.shape) - 0.5
gmm = estim_class_model_gmm(features, 4)
proba = gmm.predict_proba(features)
segment_graph_cut_general(slic, proba, gc_regul)
def estim_points_compute_features(name, img, segm, params):
""" determine points (center candidates) using slic
and for each compute feature vector with their names
:param str name:
:param ndarray img:
:param ndarray segm:
:param {str: any} params:
:return (str, ndarray, [(int, int)], [[float]], [str]):
"""
# superpixels on image
assert img.shape[:2] == segm.shape[:2], \
'not matching shapes: %r : %r' % (img.shape, segm.shape)
slic = seg_spx.segment_slic_img2d(img, params['slic_size'], params['slic_regul'])
slic_centers = seg_spx.superpixel_centers(slic)
# slic_edges = seg_spx.make_graph_segm_connect_grid2d_conn4(slic)
features, feature_names = compute_points_features(segm, slic_centers,
params)
return name, slic, slic_centers, features, feature_names
def compute_color2d_superpixels_features(image,
clr_space='rgb',
sp_size=30,
sp_regul=0.2,
dict_features=FTS_SET_SIMPLE,
fts_norm=True):
""" segment image into superpixels and estimate features per superpixel
:param ndarray image: input RGB image
:param str clr_space: chose the color space
:param int sp_size: initial size of a superpixel(meaning edge lenght)
:param float sp_regul: regularisation in range(0;1) where "0" gives elastic
and "1" nearly square segments
:param {str: [str]} dict_features: list of features to be extracted
:param bool fts_norm: weather nomalise features
:return [[int]], [[floats]]: superpixels and related of features
"""
assert sp_regul > 0., 'slic. regularisation must be positive'
logging.debug('run Superpixel clustering.')
slic = seg_sp.segment_slic_img2d(image,
sp_size=sp_size,
rltv_compact=sp_regul)
# plt.figure(), plt.imshow(slic)
logging.debug('extract slic/superpixels features.')
image = convert_img_color_space(image, clr_space)
features, _ = seg_fts.compute_selected_features_img2d(
image, slic, dict_features)
logging.debug('list of features RAW: %s', repr(features.shape))
features[np.isnan(features)] = 0
if fts_norm:
logging.debug('norm all features.')
features, _ = seg_fts.norm_features(features)
logging.debug('list of features NORM: %s', repr(features.shape))
return slic, features
def image_segmentation(idx_row, params, debug_export=DEBUG_EXPORT):
""" image segmentation which prepare inputs (imsegm, centres)
and perform segmentation of various imsegm methods
:param (int, str) idx_row: input image and centres
:param dict params: segmentation parameters
:return str: image name
"""
_, row_path = idx_row
for k in dict(row_path):
if isinstance(k, str) and k.startswith('path_'):
row_path[k] = tl_data.update_path(row_path[k], absolute=True)
logging.debug('segmenting image: "%s"', row_path['path_image'])
name = os.path.splitext(os.path.basename(row_path['path_image']))[0]
img = load_image(row_path['path_image'])
# make the image like RGB
img_rgb = np.rollaxis(np.tile(img, (3, 1, 1)), 0, 3)
seg = load_image(row_path['path_segm'], 'segm')
assert img_rgb.shape[:2] == seg.shape, \
'image %r and segm %r do not match' % (img_rgb.shape[:2], seg.shape)
if not os.path.isfile(row_path['path_centers']):
logging.warning('no center was detected for "%s"', name)
return name
centers = tl_data.load_landmarks_csv(row_path['path_centers'])
centers = tl_data.swap_coord_x_y(centers)
if not list(centers):
logging.warning('no center was detected for "%s"', name)
return name
# img = seg / float(seg.max())
slic = seg_spx.segment_slic_img2d(img_rgb,
sp_size=params['slic_size'],
relative_compact=params['slic_regul'])
path_segm = os.path.join(params['path_exp'], 'input', name + '.png')
export_draw_image_segm(path_segm, img_rgb, segm_obj=seg, centers=centers)
seg_simple = simplify_segm_3cls(seg)
path_segm = os.path.join(params['path_exp'], 'simple', name + '.png')
export_draw_image_segm(path_segm, seg_simple - 1.)
dict_segment = create_dict_segmentation(params, slic, seg, img, centers)
image_name = name + '.png'
centre_name = name + '.csv'
# iterate over segmentation methods and perform segmentation on this image
for method in dict_segment:
(fn, args) = dict_segment[method]
logging.debug(' -> %s on "%s"', method, name)
path_dir = os.path.join(params['path_exp'], method) # n.split('_')[0]
path_segm = os.path.join(path_dir, image_name)
path_centre = os.path.join(path_dir + DIR_CENTRE_POSIX, centre_name)
path_fig = os.path.join(path_dir + DIR_VISUAL_POSIX, image_name)
path_debug = os.path.join(path_dir + DIR_DEBUG_POSIX, name)
# assuming that segmentation may fail
try:
t = time.time()
if debug_export and 'rg2sp' in method:
os.mkdir(path_debug)
segm_obj, centers, dict_export = fn(*args,
debug_export=path_debug)
else:
segm_obj, centers, dict_export = fn(*args)
# also export ellipse params here or inside the segm fn
if dict_export is not None:
for k in dict_export:
export_partial(k, dict_export[k], path_dir, name)
logging.info('running time of %r on image "%s" is %d s',
fn.__name__, image_name,
time.time() - t)
tl_data.io_imsave(path_segm, segm_obj.astype(np.uint8))
export_draw_image_segm(path_fig, img_rgb, seg, segm_obj, centers)
# export also centers
centers = tl_data.swap_coord_x_y(centers)
tl_data.save_landmarks_csv(path_centre, centers)
except Exception:
logging.exception('segment fail for "%s" via %s', name, method)
return name
def segment_image(imgs_idx_path, params, classif, path_out, path_visu=None,
show_debug_imgs=SHOW_DEBUG_IMAGES):
""" perform image segmentation on input image with given paramters
and trained classifier, and save results
:param (int, str) imgs_idx_path:
:param {str: ...} params: segmentation parameters
:param obj classif: trained classifier
:param str path_out: path for output
:param str path_visu: the existing patch means export also visualisation
:return (str, ndarray, ndarray):
"""
idx, path_img = parse_imgs_idx_path(imgs_idx_path)
logging.debug('segmenting image: "%s"', path_img)
idx_name = get_idx_name(idx, path_img)
img = load_image(path_img, params['img_type'])
slic = seg_spx.segment_slic_img2d(img, sp_size=params['slic_size'],
rltv_compact=params['slic_regul'])
img = seg_pipe.convert_img_color_space(img, params.get('clr_space', 'rgb'))
features, _ = seg_fts.compute_selected_features_img2d(img, slic,
params['features'])
labels = classif.predict(features)
segm = labels[slic]
path_img = os.path.join(path_out, idx_name + '.png')
logging.debug('export segmentation: %s', path_img)
img_seg = Image.fromarray(segm.astype(np.uint8))
img_seg.convert('L').save(path_img)
# io.imsave(path_img, segm)
# plt.imsave(os.path.join(path_out, idx_name + '_rgb.png'), seg_pipe)
if path_visu is not None and os.path.isdir(path_visu):
export_draw_image_segm_contour(img, segm, path_visu, idx_name)
try: # in case some classiefier do not support predict_proba
proba = classif.predict_proba(features)
segm_soft = proba[slic]
path_npz = os.path.join(path_out, idx_name + '.npz')
np.savez_compressed(path_npz, segm_soft)
except Exception:
logging.warning('classif: %s not support predict_proba(.)',
repr(classif))
proba = None
segm_soft = None
# if probabilities was not estimated of GC regul. is zero
if proba is not None and params['gc_regul'] > 0:
gc_regul = params['gc_regul']
if params['gc_use_trans']:
label_penalty = seg_gc.compute_pairwise_cost_from_transitions(
params['label_transitions'])
gc_regul = (gc_regul * label_penalty)
labels_gc = seg_gc.segment_graph_cut_general(slic, proba, img, features,
gc_regul, edge_type=params['gc_edge_type'])
# labels_gc = seg_gc.segment_graph_cut_simple(slic, proba, gc_regul)
segm_gc = labels_gc[slic]
# relabel according classif classes
segm_gc = classif.classes_[segm_gc]
path_img = os.path.join(path_out, idx_name + '_gc.png')
logging.debug('export segmentation: %s', path_img)
img_seg_gc = Image.fromarray(segm_gc.astype(np.uint8))
img_seg_gc.convert('L').save(path_img)
# io.imsave(path_img, segm_gc)
if path_visu is not None and os.path.isdir(path_visu):
export_draw_image_segm_contour(img, segm_gc, path_visu,
idx_name, '_gc')
if show_debug_imgs:
labels_map = np.argmax(proba, axis=1)
plt.imsave(os.path.join(path_visu, idx_name + '_map.png'),
labels_map[slic])
if not segm_soft is None:
for lb in range(segm_soft.shape[2]):
uc_name = idx_name + '_gc_unary-lb%i.png' % lb
plt.imsave(os.path.join(path_visu, uc_name),
segm_soft[:, :, lb], vmin=0., vmax=1.,
cmap=plt.cm.Greens)
else:
segm_gc = np.zeros(segm.shape)
# gc.collect(), time.sleep(1)
return idx_name, segm, segm_gc