def vinterpolate(v,n=3,smoothing=1):
from atrous import smooth
L = len(v)
if smoothing > 0:
v = smooth(v, smoothing)
sp2 = ip.UnivariateSpline(np.arange(L),v, s=0)
xfit = np.linspace(0,L-1,L*n)
return sp2(xfit)
def vinterpolate(v, n=3, smoothing=1):
from atrous import smooth
L = len(v)
if smoothing > 0:
v = smooth(v, smoothing)
sp2 = ip.UnivariateSpline(np.arange(L), v, s=0)
xfit = np.linspace(0, L - 1, L * n)
return sp2(xfit)
def guess_seeds(seq, Nfirst=10, smoothing=4):
"""
automatically guess starting seeds for vessel walls
relies on the fact that vessel is the largest bright stripe
"""
Nfirst = min(len(seq), Nfirst)
d = seq[:Nfirst]
try:
from scipy.stats import skew
s = np.sign(skew(np.ravel(d)))
print 'Skewness sign:', s
except:
s = 1
y = atrous.smooth(np.mean(s * d, axis=0), smoothing)
(xfit, yfit), (mx, mn), (gups, gdowns) = lib.extrema2(y, sort_values=True)
# highest gradient up to the left of the highest max
gu1 = (g for g in gups if g < mx[0]).next()
# highest gradient up to the right of the highest max
gd1 = (g for g in gdowns if g > mx[0]).next()
return (xfit[gu1], xfit[gd1])
def guess_seeds(seq, Nfirst=10,smoothing=4):
"""
automatically guess starting seeds for vessel walls
relies on the fact that vessel is the largest bright stripe
"""
Nfirst = min(len(seq), Nfirst)
d = seq[:Nfirst]
try:
from scipy.stats import skew
s = np.sign(skew(np.ravel(d)))
print 'Skewness sign:', s
except :
s = 1
y = atrous.smooth(np.mean(s*d,axis=0), smoothing)
(xfit,yfit), (mx,mn), (gups,gdowns) = lib.extrema2(y, sort_values=True)
# highest gradient up to the left of the highest max
gu1 = (g for g in gups if g < mx[0]).next()
# highest gradient up to the right of the highest max
gd1 = (g for g in gdowns if g > mx[0]).next()
return (xfit[gu1], xfit[gd1])
def locextr(v, x=None, mode='max', refine=10, output='xfit'):
"""Finds local extrema, type of extrema depends on parameter "mode"
mode can be {'max' | 'min' | 'gup' | 'gdown' | 'gany'}"""
if isinstance(x, str):
mode = x
if x is None or isinstance(x, str):
x = np.arange(len(v))
sp0 = ip.UnivariateSpline(x, atrous.smooth(v), s=0)
if mode in ['max', 'min']:
sp = sp0.derivative(1)
elif mode in ['gup', 'gdown', 'gany']:
sp = sp0.derivative(2)
res = 0.05
if refine > 1:
xfit = np.linspace(0, x[-1], len(x) * refine)
else:
xfit = x
di = sp(xfit)
if mode in ['max', 'gup']:
dersign = np.sign(di)
elif mode in ['min', 'gdown']:
dersign = -np.sign(di)
locations = dersign[:-1] - dersign[1:] > 1.5
if output is 'all':
out = xfit[locations], sp0(xfit)[locations]
elif output is 'yfit':
out = di[locations]
elif output is 'xfit':
out = xfit[locations]
else:
print """unknown output code, should be one of 'xfit', 'yfit', 'all',
returning 'x' locations"""
out = xfit[locations]
return out
def locextr(v, x=None, mode = 'max', refine=10, output='xfit'):
"""Finds local extrema, type of extrema depends on parameter "mode"
mode can be {'max' | 'min' | 'gup' | 'gdown' | 'gany'}"""
if type(x) is str:
mode = x
if x is None or type(x) is str:
x = np.arange(len(v))
sp0 = ip.UnivariateSpline(x,atrous.smooth(v),s=0)
if mode in ['max', 'min']:
sp = sp0.derivative(1)
elif mode in ['gup', 'gdown', 'gany']:
sp = sp0.derivative(2)
res = 0.05
if refine > 1:
xfit = np.linspace(0,x[-1], len(x)*refine)
else:
xfit = x
di = sp(xfit)
if mode in ['max', 'gup']:
dersign = np.sign(di)
elif mode in ['min', 'gdown']:
dersign = -np.sign(di)
locations = dersign[:-1] - dersign[1:] > 1.5
if output is 'all':
out = xfit[locations], sp0(xfit)[locations]
elif output is 'yfit':
out = di[locations]
elif output is 'xfit':
out = xfit[locations]
else:
print """unknown output code, should be one of 'xfit', 'yfit', 'all',
returning 'x' locations"""
out = xfit[locations]
return out
def decompose_mwt(arr,level, s=2, tau=5, upto=2):
"""Median-Wavelet transform (up to level 'upto', then followed by starlet transform"""
out = []
cj = arr
upto = min(level, upto)
for j in range(upto):
approxm = median_filter(cj, 2*s+1)
wm = cj-approxm
#th = tau*MAD(wm)/0.6745
th = tau*MAD(wm[wm!=0])/0.6745
wm[np.abs(wm) > th] = 0
#wm *= np.abs(wm) <= th
approx_dash = wm + approxm
cjp1 = atrous.smooth(approx_dash, j+1)
out.append(cj-cjp1)
cj = cjp1
s *= 2
out.append(cj)
if level > upto:
wcoefs = atrous.decompose(cj, level)
out[-1] = np.sum(wcoefs[:upto+1], axis=0)
out = np.concatenate([out, wcoefs[upto+1:]])
return out
def decompose_mwt(arr, level, s=2, tau=5, upto=2):
"""Median-Wavelet transform (up to level 'upto', then followed by starlet transform"""
out = []
cj = arr
upto = min(level, upto)
for j in range(upto):
approxm = median_filter(cj, 2 * s + 1)
wm = cj - approxm
# th = tau*MAD(wm)/0.6745
th = tau * MAD(wm[wm != 0]) / 0.6745
wm[np.abs(wm) > th] = 0
# wm *= np.abs(wm) <= th
approx_dash = wm + approxm
cjp1 = atrous.smooth(approx_dash, j + 1)
out.append(cj - cjp1)
cj = cjp1
s *= 2
out.append(cj)
if level > upto:
wcoefs = atrous.decompose(cj, level)
out[-1] = np.sum(wcoefs[: upto + 1], axis=0)
out = np.concatenate([out, wcoefs[upto + 1 :]])
return out
def v2grads(v):
L = len(v)
sp2 = ip.UnivariateSpline(np.arange(L), atrous.smooth(v), s=0)
xfit = np.arange(0, L, 0.1)
return np.abs(sp2.derivative(1)(xfit))
def v2grads(v):
L = len(v)
sp2 = ip.UnivariateSpline(np.arange(L),atrous.smooth(v), s=0)
xfit = np.arange(0,L,0.1)
return np.abs(sp2.derivative(1)(xfit))
