When reconstructing the image from the decomposition, some of the frequency information appears to be missing. The original code returns the real portion of the inverse DFT of outdft but when returning the raw outdft, some of the frequency info appears to be missing. See example below, this seems to happen for all images.

from skimage.data import camera

test_image = camera()

s = Steerable(5)
coeff, lo0mask, hi0mask = s.buildSCFpyr(test_image) 
resdft = s.reconSCFpyr(coeff) # modified to return resdft

plt.imshow(np.log(np.abs(np.fft.fftshift(np.fft.fft2(test_image))))) 
plt.imshow(np.log(np.abs(resdft))) 

Hi Dr. Dzung,

Thanks a lot for your code! Can I know what's the filter you use for steerable decomposition?

Thanks!

In filterbank.py, at line 152:

lostart = np.ceil((dims+0.5)/2) - np.ceil((np.ceil((dims-0.5)/2)+0.5)/2)
loend = lostart + np.ceil((dims-0.5)/2)

nlog_rad = log_rad[lostart[0]:loend[0], lostart[1]:loend[1]]

Variables lostart and loend are both of float type. Using their values to slice into nlog_rad causes the following:

packages/perceptual/filterbank.py", line 155, in reconSCFpyrLevs
    nlog_rad = log_rad[lostart[0]:loend[0], lostart[1]:loend[1]]
TypeError: slice indices must be integers or None or have an __index__ method

in
./perceptual/filterbanl.py
line 241

sc.factorial

AttributeError: module 'scipy.misc' has no attribute 'factorial'

can be changed to sc.special.factorial

can see this

nipy/nipy#453