def phase_congruency(im,
N,
w,
sigma,
scales,
scale_factor=2,
model=None,
model_factor=1):
plt.matshow(im)
plt.show()
if np.ndim(im) > 2:
im = skc.rgb2gray(im)
ch = np.zeros(im.shape + (2 * N + 1, scales), dtype=np.complex)
for i in range(scales):
spectrum = log_gabor_spectrum(im.shape[:2],
w * scale_factor**i,
sigma,
passband='band')
ch_i = img_chv(im, spectrum, N)
if model is not None:
_, ch_i, _ = create_poly(ch_i, model)
# print(ch_i[:,400, 400, :])
ch_i = ch_i[0] + 1j * ch_i[1]
# print(ch_i[400,400,:])
# ch_i = ch_i[:, :, ::model_factor]
ch[:, :, :, i] = ch_i
# plt.matshow(chv_norm(ch[:, :, :, i]))
# plt.show()
A1 = np.linalg.norm(np.sum(ch, axis=-1), axis=-1)
A2 = np.sum(np.linalg.norm(ch, axis=-2).squeeze(), axis=-1)
# plt.matshow(A1)
# plt.colorbar()
# plt.show()
# plt.matshow(A2)
# plt.colorbar()
# plt.show()
epsilon = 0.01
plt.matshow(A1 / (A2 + epsilon))
plt.colorbar()
plt.show()
plt.matshow(beta.cdf(A1 / (A2 + epsilon), 5, 1))
plt.colorbar()
plt.show()
return A1 / (A2 + epsilon)
def scale(im, N, w, sigma, scales, scale_factor=2):
if np.ndim(im) > 2:
im = skc.rgb2gray(im)
ch = np.zeros(im.shape + (2 * N + 1, scales), dtype=np.complex)
A = []
for i in range(scales):
spectrum = log_gabor_spectrum(im.shape[:2],
w * scale_factor**i,
sigma,
passband='band')
ch_i = img_chv(im, spectrum, N)
A.append(chv_norm(ch_i))
A = np.asarray(A)
return np.max(A, axis=0), np.argmax(A, axis=0)
sigma = 0.65
# CH vector weights
weightV = weights.sinusoid_weights(N, 0, 0)
print("Weights are: ")
print(weightV.transpose())
# CH vector output array
ch = np.zeros(ch_shape, dtype=np.complex)
# Calculate
for i, w in enumerate(wavelengths):
print("Calculating CH vector for wavelength {}".format(w))
if i == 0:
passband = 'high'
else:
passband = 'band'
gf = log_gabor_spectrum(im.shape, w, sigma)
ch[i] = chv.img_chv(im, gf, N, weightV)
# Calculate model parameters
# Model vectors
U = sinusoid_pair(N, weightV)
# Test
# cht = U[1]
# cht = cht[np.newaxis, np.newaxis, :]
# poly, delt, lamb = create_poly(cht, U)
# poly = poly[:, :, ::2]
# roots = poly_roots(poly)
# th = np.angle(roots)
# val_at_roots = poly_value(poly, th)
# max_idx = np.argmax(val_at_roots, axis=2)
# th_max = np.take_along_axis(th, max_idx[:, :, np.newaxis], axis=2)
import skimage.io as skio
import skimage.color as skcolor
import matplotlib.pyplot as plt
import colorcet as cc
import chvector.transforms.chv as chv
from chvector.filters import log_gabor_spectrum
from chvector.models.archetypes import *
from chvector.models.solver import *
s = sinusoid_pair(7)
print(s)
im = skcolor.rgb2gray(skio.imread("COB3-orig.png"))
skio.imshow(im)
plt.show()
plt.pause(0)
gf = log_gabor_spectrum(im.shape[:2], 4, 0.5)
ch = chv.img_chv(im, gf, 7)
poly = create_poly(ch, s)
mag = np.linalg.norm(ch, axis=2)
plt.imshow(np.log(mag), cmap=cc.cm.fire)
plt.show()
plt.pause(0)
sigma = 0.65
# CH vector weights
wv = weights.sinusoid_weights(N, 0, 0)
print("Weights are: ")
print(wv.transpose())
# CH vector output array
ch = np.zeros(im.shape + (2 * N + 1, len(wavelengths)), dtype=np.complex)
# Calculate
for i, w in enumerate(wavelengths):
print("Calculating CH vector for wavelength {}".format(w))
if i == 0:
passband = 'high'
else:
passband = 'band'
gf = log_gabor_spectrum(im.shape, w, sigma)
ch[:, :, :, i] = chv.img_chv(im, gf, N, wv)
# Calculate model parameters
# Model vectors
U = sinusoid_pair(N, w)
# Model output array
A_model = np.zeros(im.shape + (len(wavelengths), ))
phi_model = np.zeros(im.shape + (len(wavelengths), ))
theta_model = np.zeros(im.shape + (len(wavelengths), ))
# Calculate
for i in range(ch.shape[3]):
# Create response polynomials
poly, lamb, delt = create_poly(ch[:, :, :, i], U)
# Solve for maximum
# Sinusoid has poly order of 2
poly = poly[:, :, ::2]