def process(self, rgb_img):
img = utils.image.rgb2ihls(rgb_img)
f = {}
channels = {'H' : img[:,:,0], 'S' : img[:,:,1], 'Y' : img[:,:,2]}
# For each channel
for c, img in channels.items() :
# Initiate aggregate over all levels
f[c+"_sum"] = 0.0
# For each wavelet level
for l in xrange(3) :
dimg = mahotas.daubechies(img, 'D8')
h, w = dimg.shape
mh, mw = h/2, w/2
w_lh = np.mean(np.abs(dimg[:mh,mw:]))
w_hh = np.mean(np.abs(dimg[mh:,mw:]))
w_hl = np.mean(np.abs(dimg[mh:,:mw]))
# Store the value in the correct name and sum the aggregate over all levels
w = (w_lh + w_hl + w_hh)/3
f[c+str(l)] = w
f[c+"_sum"] += w
img = img[:mh,:mw]
return f
def test_daubechies_idaubechies():
f = luispedro_jpg(1)
f = f[:256,:256]
fo = f.copy()
d = mahotas.daubechies(f, 'D8')
r = mahotas.idaubechies(d, 'D8')
assert np.mean( (r[4:-4,4:-4] - fo[4:-4,4:-4])**2) < 1.
def test_wavelets_inline():
def inline(f):
im = np.arange(16, dtype=float).reshape((4,4))
t = f(im, inline=True)
assert id(im) == id(t)
yield inline, mahotas.haar
yield inline, lambda im,inline: mahotas.daubechies(im, 'D4', inline=inline)
def test_3d_wavelets_error():
@raises(ValueError)
def call_f(f):
f(np.arange(4*4*4).reshape((4,4,4)))
yield call_f, mahotas.haar
yield call_f, mahotas.ihaar
yield call_f, lambda im: mahotas.daubechies(im, 'D4')
def test_daubechies_D2_haar():
image = luispedro_jpg(1)
image = image[:256,:256]
wav = mahotas.haar(image, preserve_energy=False)
dau = mahotas.daubechies(image, 'D2')
assert wav.shape == dau.shape
assert np.allclose(dau, wav)
def test_daubechies_idaubechies():
f = luispedro_jpg()
f = f[:256, :256]
fo = f.copy()
d = mahotas.daubechies(f, 'D8')
r = mahotas.idaubechies(d, 'D8')
assert np.mean((r[4:-4, 4:-4] - fo[4:-4, 4:-4])**2) < 1.
def test_3d_wavelets_error():
@raises(ValueError)
def call_f(f):
f(np.arange(4 * 4 * 4).reshape((4, 4, 4)))
yield call_f, mahotas.haar
yield call_f, mahotas.ihaar
yield call_f, lambda im: mahotas.daubechies(im, 'D4')
def test_daubechies_D2_haar():
image = luispedro_jpg()
image = image[:256, :256]
wav = mahotas.haar(image, preserve_energy=False)
dau = mahotas.daubechies(image, 'D2')
assert wav.shape == dau.shape
assert np.allclose(dau, wav)
def test_wavelets_inline():
def inline(f):
im = np.arange(16, dtype=float).reshape((4, 4))
t = f(im, inline=True)
assert id(im) == id(t)
yield inline, mahotas.haar
yield inline, lambda im, inline: mahotas.daubechies(
im, 'D4', inline=inline)
def test_center_wavelet_iwavelet_decenter():
from mahotas import wavelet_center, wavelet_decenter
import mahotas
import numpy as np
f = luispedro_jpg(1)
f = f[:100,:250]
fo = f.copy()
for wav in ('D2', 'D4', 'D6', 'D8', 'D10', 'D12', 'D16'):
fc = mahotas.wavelet_center(fo, border=24)
t = mahotas.daubechies(fc, wav)
r = mahotas.idaubechies(t, wav)
rd = mahotas.wavelet_decenter(r, fo.shape, border=24)
assert np.allclose(fo, rd)
def test_center_wavelet_iwavelet_decenter():
from mahotas import wavelet_center, wavelet_decenter
import mahotas
import numpy as np
f = luispedro_jpg()
f = f[:100, :250]
fo = f.copy()
for wav in ('D2', 'D4', 'D6', 'D8', 'D10', 'D12', 'D16'):
fc = mahotas.wavelet_center(fo, border=24)
t = mahotas.daubechies(fc, wav)
r = mahotas.idaubechies(t, wav)
rd = mahotas.wavelet_decenter(r, fo.shape, border=24)
assert np.allclose(fo, rd)
def test_non_valid_daubechies():
image = luispedro_jpg()
mahotas.daubechies(image, 'D-4')
def test_non_valid_daubechies():
image = luispedro_jpg()
mahotas.daubechies(image, 'D-4')
f = f[:256, :256]
plt.gray()
# Show the data:
plt.imshow(f)
print("Fraction of zeros in original image:", np.mean(f == 0))
# A baseline compression method: save every other pixel and only high-order bits:
direct = f[::2, ::2].copy()
direct /= 8
direct = direct.astype(np.uint8)
print("Fraction of zeros in original image (after division by 8):",
np.mean(direct == 0))
plt.imshow(direct)
# Transform using D8 Wavelet to obtain transformed image t:
t = mahotas.daubechies(f, 'D8')
plt.imshow(t)
# Discard low-order bits:
t /= 8
t = t.astype(np.int8)
print("Fraction of zeros in transform (after division by 8):", np.mean(t == 0))
plt.imshow(t)
# Let us look at what this looks like
r = mahotas.idaubechies(t, 'D8')
plt.imshow(r)
# Go further, discard small values in the transformed space:
tt = soft_threshold(t, 12)
print(
f = f[:256,:256]
plt.gray()
# Show the data:
plt.imshow(f)
print("Fraction of zeros in original image:", np.mean(f==0))
# A baseline compression method: save every other pixel and only high-order bits:
direct = f[::2,::2].copy()
direct /= 8
direct = direct.astype(np.uint8)
print("Fraction of zeros in original image (after division by 8):", np.mean(direct==0))
plt.imshow(direct)
# Transform using D8 Wavelet to obtain transformed image t:
t = mahotas.daubechies(f,'D8')
plt.imshow(t)
# Discard low-order bits:
t /= 8
t = t.astype(np.int8)
print("Fraction of zeros in transform (after division by 8):", np.mean(t==0))
plt.imshow(t)
# Let us look at what this looks like
r = mahotas.idaubechies(t, 'D8')
plt.imshow(r)
# Go further, discard small values in the transformed space:
tt = soft_threshold(t, 12)
print("Fraction of zeros in transform (after division by 8 & soft thresholding):", np.mean(tt==0))
