def segment_fit_ellipse(seg,
centers,
fn_preproc_points,
thr_overlap=SEGM_OVERLAP):
""" segment eggs using ellipse fitting
:param ndarray seg: input image / segmentation
:param [[int, int]] centers: position of centres / seeds
:param fn_preproc_points: function for detection boundary points
:param float thr_overlap: threshold for removing overlapping segmentation
:return ndarray, [[int, int]]: resulting segmentation, updated centres
"""
points_centers = fn_preproc_points(seg, centers)
centres_new, ell_params = [], []
segm = np.zeros_like(seg)
for i, points in enumerate(points_centers):
lb = i + 1
ellipse = EllipseModel()
ellipse.estimate(points)
if ellipse is None:
continue
logging.debug('ellipse params: %s', repr(ellipse.params))
segm = ell_fit.add_overlap_ellipse(segm, ellipse.params, lb,
thr_overlap)
if np.any(segm == lb):
centres_new.append(centers[i])
ell_params.append(ellipse.params)
dict_export = {
'ellipses.csv':
pd.DataFrame(ell_params, columns=['xc', 'yc', 'a', 'b', 'theta'])
}
return segm, np.array(centres_new), dict_export
def segment_fit_ellipse_ransac(seg, centers, fn_preproc_points, nb_inliers=0.6,
thr_overlap=SEGM_OVERLAP):
""" segment eggs using ellipse fitting and RANDSAC strategy
:param ndarray seg: input image / segmentation
:param [[int, int]] centers: position of centres / seeds
:param fn_preproc_points: function for detection boundary points
:param float nb_inliers: ratio of inliers for RANSAC
:param float thr_overlap: threshold for removing overlapping segmentations
:return ndarray, [[int, int]]: resulting segmentation, updated centres
"""
points_centers = fn_preproc_points(seg, centers)
centres_new = []
segm = np.zeros_like(seg)
for i, points in enumerate(points_centers):
lb = i + 1
nb_min = int(len(points) * nb_inliers)
ransac_model, _ = measure.ransac(points, EllipseModel,
min_samples=nb_min,
residual_threshold=15,
max_trials=250)
if ransac_model is None:
continue
logging.debug('ellipse params: %s', repr(ransac_model.params))
segm = ell_fit.add_overlap_ellipse(segm, ransac_model.params, lb,
thr_overlap)
if np.any(segm == lb):
centres_new.append(centers[i])
return segm, np.array(centres_new)
def segment_fit_ellipse_ransac_segm(seg,
centers,
fn_preproc_points,
table_p,
nb_inliers=0.35,
thr_overlap=SEGM_OVERLAP):
""" segment eggs using ellipse fitting and RANDSAC strategy on segmentation
:param ndarray seg: input image / segmentation
:param [[int, int]] centers: position of centres / seeds
:param fn_preproc_points: function for detection boundary points
:param [[float]] table_p: table of probabilities being foreground / background
:param float nb_inliers: ratio of inliers for RANSAC
:param float thr_overlap: threshold for removing overlapping segmentations
:return ndarray, [[int, int]]: resulting segmentation, updated centres
"""
slic, points_all, labels = ell_fit.get_slic_points_labels(seg,
slic_size=15,
slic_regul=0.1)
points_centers = fn_preproc_points(seg, centers)
weights = np.bincount(slic.ravel())
centres_new, ell_params = [], []
segm = np.zeros_like(seg)
for i, points in enumerate(points_centers):
lb = i + 1
ransac_model, _ = ell_fit.ransac_segm(points,
ell_fit.EllipseModelSegm,
points_all,
weights,
labels,
table_p,
min_samples=nb_inliers,
residual_threshold=25,
max_trials=250)
if ransac_model is None:
continue
logging.debug('ellipse params: %s', repr(ransac_model.params))
segm = ell_fit.add_overlap_ellipse(segm, ransac_model.params, lb,
thr_overlap)
if np.any(segm == lb):
centres_new.append(centers[i])
ell_params.append(ransac_model.params)
dict_export = {
'ellipses.csv':
pd.DataFrame(ell_params, columns=['xc', 'yc', 'a', 'b', 'theta'])
}
return segm, np.array(centres_new), dict_export