scipy.signal.wiener always show warning

Question

I'm trying to use the wiener denoise function from Scipy in a grayscale image:

image = data.imread(image_path, as_grey=True)
image = signal.wiener(image)

but I always get this warning:

/usr/lib/python2.7/dist-packages/scipy/signal/signaltools.py:425: RuntimeWarning: divide by zero encountered in true_divide
      res *= (1 - noise / lVar)
    /usr/lib/python2.7/dist-packages/scipy/signal/signaltools.py:425: RuntimeWarning: invalid value encountered in multiply
      res *= (1 - noise / lVar)

The filter works fine, but why I get this warning?

I can reproduce those warnings if all the values in `image` are the same, e.g. `wiener(np.ones((24, 24)))`. Is that the case with your input? — Warren Weckesser, Oct 29 '15 at 14:48
Actually, the array doesn't have to be constant. I also get warnings with `wiener(np.random.randint(0, 2, size=(24, 24))`. — Warren Weckesser, Oct 29 '15 at 14:55
This bug looks to be related: https://github.com/scipy/scipy/issues/9015 — Diarmaid O Cualain, Jul 22 '22 at 09:39

score 1 · Accepted Answer · answered Feb 19 '16 at 03:25

1

Try to convert the data type of input image to float32 or float64, that is:

image = image.astype('float64')
image = signal.wiener(image)

I guess we countered the "divided by zero" as the underflow number.

answered Feb 19 '16 at 03:25

wllbll

531
5
11

scipy.signal.wiener always show warning

1 Answers1