def find_testlightFromRG(self, r, g):
'''Returns -1 if not found (extra spectral or doesn't live on locus)
'''
err = lambda r, g, lam: np.sqrt((r - self.rVal[lam]) ** 2 + (g -
self.gVal[lam]) ** 2)
# Check if point is extra spectral
a = np.array([r, g]) - np.array([self.rVal[0], self.gVal[0]])
b = np.array([self.rVal[-1], self.gVal[-1]]) - np.array(
[self.rVal[0], self.gVal[0]])
a = int(cart2pol(a[0], a[1])[0])
b = int(cart2pol(b[0], b[1])[0])
if a == b:
return -1 # This point is extra spectral
i = 0
startE = err(r, g, i)
forward = True
while forward:
e = err(r, g, i)
if startE < e and e < 0.05:
forward = False
else:
startE = e
if i + 1 > len(self.spectrum) - 1:
return -1 # either extra spectral or not on spec locus
i += 1
t0 = err(r, g, i)
t1 = err(r, g, i - 1)
e1 = t1 / (t1 + t0)
# take a weighted averge of the points
outLam = self.spectrum[i] * e1 + self.spectrum[i - 1] * (1 - e1)
return outLam
def find_dominant_wl(self, rg, white=[1/3, 1/3]):
'''Find the dominant wavelength from a given point within the
chromaticity diagram. If extra spectral, returns str.
'''
rg = np.asarray(rg)
white = np.asarray(white)
## make sure both are len 3 vectors
if len(rg) == 2:
rgb = np.array([rg[0], rg[1], 1 - (rg.sum())])
else:
rgb = rg
if len(white) == 2:
white = np.array([white[0], white[1], 1 - white.sum()])
point = rgb - white
# make sure white point wasn't fed in
if round(point[0], 2) == 0.0 and round(point[1] == 0.0, 2):
return 0
neutral_points = self.find_spect_neutral(rgb, white)
dom = None
for i, neutral in enumerate(neutral_points):
n = neutral - white[:2]
a = int(cart2pol(n[0], n[1])[0])
b = int(cart2pol(point[0], point[1])[0])
if a == b:
dom = self.find_testlightFromRG(neutral[0], neutral[1])
break
# must be extra spectral: use complementary color
if dom == -1:
if i == 0:
i = 1
if i == 1:
i = 0
neutral = neutral_points[i]
# return as negative number to indicate it is a complement
dom = -1 * self.find_testlightFromRG(neutral[0], neutral[1])
elif dom == None:
raise ValueError('dominant wavelength not found')
return dom
def welch2d(x,L = None, over = 0.5, win = 2.0):
"""2D spectrum estimation using Welch's method.
The spectrum of an image x is estimated using Welch's \
method of averaging modified periodograms.
:param x: image
:param L: section length
:param over: amount of overlap, where 0<over<1,
:param win: The window type \n
1 = Rectangular \n
2 = Hamming \n
3 = Hanning \n
4 = Bartlett \n
5 = Blackman \n
:returns: Welch's estimate of the power spectrum, returned in decibels.
.. note::
Modified from: M.H. Hayes. \
"Statistical Digital Signal Processing and Modeling" \
(John Wiley & Sons, 1996).
"""
xs, ys = x.shape
x = RemoveMean(x)
if L == None:
L = xs
if xs / ys != 1.0:
raise ValueError('This is a stupid program. Dimensions need to be \
equal (len(x)=len(y))')
if L < len(x[:,0]) / 2.0:
raise ValueError('Length must be longer than 1/2 length of x')
if (over >= 1) or (over < 0):
raise ValueError('Overlap is invalid')
n0 = (1.0 - over) * L
n1 = np.array([1.0, 1.0]) - n0
n2 = np.array([L, L]) - n0
nsect = int(1.0 + np.floor((len(x) - L) /( n0)))
Px = 0
for ix in range(0,nsect):
n1[0] = n1[0] + n0
n2[0] = n2[0] + n0
for iy in range(0,nsect):
n1[1] = n1[1] + n0
n2[1] = n2[1] + n0
Px += PowerSpectrum2(x[n1[0]:n2[0],n1[1]:n2[1]], win) / (nsect**2)
xs, ys = Px.shape
f2 = np.arange(-xs / 2.0, xs / 2.0)
f1 = np.arange(-ys / 2.0, ys / 2.0)
XX, YY = np.meshgrid(f1,f2)
foo, r = dm.cart2pol(XX,YY)
if np.mod(xs,2)==1 or np.mod(ys,2)==1:
r = np.around(r)-1
else:
r = np.around(r)
r = np.array(r,dtype=int)
### need to possibly use a for loop.
avg = dm.accum(r.flatten() + 1, Px.flatten()
)[1:] / dm.accum(r.flatten() + 1)[1:]
avg = avg[:(xs / 2 + 1)]
return avg
