def symmetrize(mask, cx, cy):
_mask = mask.copy()
_mask *= _turn180(_mask, cx, cy)
return _mask
def symmetrize(mask,cx,cy):
_mask = mask.copy()
_mask *= _turn180(_mask,cx,cy)
return _mask
def find_center_pixelwise_slow(img, msk, x0=0, y0=0, dmax=5, rmax=None):
"""
Find center of diffraction pattern using pixelwise comparison of centro-symmetric pixels.
This is the original python implementation taken from owl/fit.py
usage:
======
x,y = find_center_pixelwise_slow(img, msk, x0, y0, dmax=5, rmax=None)
"""
s = img.shape
if rmax is None: rmax = np.sqrt(2)*max(s)
cx_g = (s[1]-1)/2.+x0
cy_g = (s[0]-1)/2.+y0
cx_g = np.round(cx_g*2)/2.
cy_g = np.round(cy_g*2)/2.
ddc = 0.5
N_sam1= int(np.round(2*dmax/ddc))+1
cx_sam1 = np.linspace(cx_g-dmax,cx_g+dmax,N_sam1)
cy_sam1 = np.linspace(cy_g-dmax,cy_g+dmax,N_sam1)
N_sam2= int(np.round(4*dmax/ddc))+1
cx_sam2 = np.linspace(cx_g-dmax*2,cx_g+dmax*2,N_sam2)
cy_sam2 = np.linspace(cy_g-dmax*2,cy_g+dmax*2,N_sam2)
# extend mask so that every pixel has a partner at every possible center
msk_ext = msk.copy()
for cy in cy_sam2:
for cx in cx_sam2:
msk_ext *= _symmetrize(msk,cx,cy)
Nme = msk_ext.sum()
errs = np.zeros(shape=(N_sam1,N_sam1))
r_max_sq = rmax**2
X,Y = np.meshgrid(np.arange(img.shape[1]),np.arange(img.shape[0]))
for cx,icx in zip(cx_sam1,np.arange(N_sam1)):
for cy,icy in zip(cy_sam1,np.arange(N_sam1)):
# SLOW CODE
#for x,y,v1 in zip((Xme-cx),(Yme-cy),imgme):
# M = (Xm-cx==-x)*(Ym-cy==-y)
# if M.sum() == 1:
# v2 = imgm[M==True]
# errs[icy,icx] += abs(v1-v2)
# else:
# print x,y,M.sum()
# FAST CODE (does the same)
r_sq = ((X-cx)**2+(Y-cy)**2)
rmsk = r_sq < r_max_sq
img_turned = _turn180(img,cx,cy)
diff = abs((img-img_turned)*msk_ext*rmsk)
errs[icy,icx] = diff.sum()
#print cx,cy,errs[icy,icx]
errs_sm = _gaussian_smooth_2d1d(errs,dmax)
i_min = errs.flatten().argmin()
cxi_min = i_min % N_sam1
cyi_min = i_min/N_sam1
#if full_output:
# D = {}
# D["msk_ext"] = msk_ext
# D["img_turned_msk_ext"] = img_turned*msk_ext
# D["img_msk_ext"] = img*msk_ext
# D["errmap"] = errs/errs.max()
# D["errmap_sm"] = gaussian_smooth_2d1d(errs,3.)
# D["errmap_sm"] /= D["errmap_sm"].max()
# D["errmap_min"] = errs.min()
# return cx_sam1[cxi_min],cy_sam1[cyi_min],D
#else:
cx_r = cx_sam1[cxi_min]
cy_r = cy_sam1[cyi_min]
x = cx_r-(s[1]-1)/2.
y = cy_r-(s[0]-1)/2.
return (x,y, errs.flatten()[i_min])
def find_center_pixelwise_slow(img, msk, x0=0, y0=0, dmax=5, rmax=None):
"""
Find center of diffraction pattern using pixelwise comparison of centro-symmetric pixels.
This is the original python implementation taken from owl/fit.py
usage:
======
x,y = find_center_pixelwise_slow(img, msk, x0, y0, dmax=5, rmax=None)
"""
s = img.shape
if rmax is None: rmax = np.sqrt(2)*max(s)
cx_g = (s[1]-1)/2.+x0
cy_g = (s[0]-1)/2.+y0
cx_g = np.round(cx_g*2)/2.
cy_g = np.round(cy_g*2)/2.
ddc = 0.5
N_sam1= int(np.round(2*dmax/ddc))+1
cx_sam1 = np.linspace(cx_g-dmax,cx_g+dmax,N_sam1)
cy_sam1 = np.linspace(cy_g-dmax,cy_g+dmax,N_sam1)
N_sam2= int(np.round(4*dmax/ddc))+1
cx_sam2 = np.linspace(cx_g-dmax*2,cx_g+dmax*2,N_sam2)
cy_sam2 = np.linspace(cy_g-dmax*2,cy_g+dmax*2,N_sam2)
# extend mask so that every pixel has a partner at every possible center
msk_ext = msk.copy()
for cy in cy_sam2:
for cx in cx_sam2:
msk_ext *= _symmetrize(msk,cx,cy)
Nme = msk_ext.sum()
errs = np.zeros(shape=(N_sam1,N_sam1))
r_max_sq = rmax**2
X,Y = np.meshgrid(np.arange(img.shape[1]),np.arange(img.shape[0]))
for cx,icx in zip(cx_sam1,np.arange(N_sam1)):
for cy,icy in zip(cy_sam1,np.arange(N_sam1)):
# SLOW CODE
#for x,y,v1 in zip((Xme-cx),(Yme-cy),imgme):
# M = (Xm-cx==-x)*(Ym-cy==-y)
# if M.sum() == 1:
# v2 = imgm[M==True]
# errs[icy,icx] += abs(v1-v2)
# else:
# print x,y,M.sum()
# FAST CODE (does the same)
r_sq = ((X-cx)**2+(Y-cy)**2)
rmsk = r_sq < r_max_sq
img_turned = _turn180(img,cx,cy)
diff = abs((img-img_turned)*msk_ext*rmsk)
errs[icy,icx] = diff.sum()
#print cx,cy,errs[icy,icx]
errs_sm = _gaussian_smooth_2d1d(errs,dmax)
i_min = errs.flatten().argmin()
cxi_min = i_min % N_sam1
cyi_min = i_min/N_sam1
#if full_output:
# D = {}
# D["msk_ext"] = msk_ext
# D["img_turned_msk_ext"] = img_turned*msk_ext
# D["img_msk_ext"] = img*msk_ext
# D["errmap"] = errs/errs.max()
# D["errmap_sm"] = gaussian_smooth_2d1d(errs,3.)
# D["errmap_sm"] /= D["errmap_sm"].max()
# D["errmap_min"] = errs.min()
# return cx_sam1[cxi_min],cy_sam1[cyi_min],D
#else:
cx_r = cx_sam1[cxi_min]
cy_r = cy_sam1[cyi_min]
x = cx_r-(s[1]-1)/2.
y = cy_r-(s[0]-1)/2.
return (x,y, errs.flatten()[i_min])
