Re: [scikit-image] Observed difference between skimage.filters.rank.mean and scipy.signal.convolve2d
Hi Victor, The biggest problem is that you’re getting bitten by datatypes. Please read the following document: http://scikit-image.org/docs/dev/user_guide/data_types.html Specifically, rank_mean is a uint8 image, so only contains integers between 0 and 255. naive_convolve is a float image, with continuous values between 0 and 255. Try this: In [17]: naive_convolve_float = naive_convolve / 255 In [18]: rank_mean_float = rank_mean / 255 In [19]: plt.imshow(np.abs(naive_convolve_float - rank_mean_float), cmap='magma') Out[19]: <matplotlib.image.AxesImage at 0x10d098fd0> Result: There is a smaller difference also. Internally, filters.rank uses a fancy rolling histogram algorithm with integer data values. This means that the result of the rank_mean is only approximately accurate, essentially to within integer rounding (good enough for most real-world uses), while the convolve2d code gives you an exact value (to within floating point error). Hope this helps! Juan. On 2 Nov 2017, 7:54 PM +1100, Poughon Victor <Victor.Poughon@cnes.fr>, wrote:
Hello,
I looks like skimage.filters.rank.mean and scipy.signal.convolve2d don't output exactly the same images. When doing:
image = data.coins() K = np.ones((11, 11))
rank_mean = rank.mean(image, selem=K) naive_convolve = convolve2d(image, K, mode="same") / K.sum()
All output pixel are different, with an absolute difference varying randomly between 0 and 1. Of course there's also a massive difference at the border, but that's expected because convolve2d treats image boundaries differently. But even in the center of the image all pixels are different. I've made a test script with an illustrated output image, you can check it out in this gist:
https://gist.github.com/vpoughon/b4afc76ce5dc681fda9d0550d41359d3
Am I doing something wrong?
Thanks,
Victor Poughon
_______________________________________________ scikit-image mailing list scikit-image@python.org https://mail.python.org/mailman/listinfo/scikit-image
Thank you, that makes sense. Your link does not explain why float images are restricted to [-1; 1] though? So with my data I can't just do: image = np.array(data.coins(), dtype=np.float) I guess skimage makes an assumption that images are within a fixed range? Best, Victor Poughon ________________________________ De : scikit-image [scikit-image-bounces+victor.poughon=cnes.fr@python.org] de la part de Juan Nunez-Iglesias [jni.soma@gmail.com] Envoyé : vendredi 3 novembre 2017 14:56 À : Mailing list for scikit-image (http://scikit-image.org) Objet : Re: [scikit-image] Observed difference between skimage.filters.rank.mean and scipy.signal.convolve2d Hi Victor, The biggest problem is that you’re getting bitten by datatypes. Please read the following document: http://scikit-image.org/docs/dev/user_guide/data_types.html Specifically, rank_mean is a uint8 image, so only contains integers between 0 and 255. naive_convolve is a float image, with continuous values between 0 and 255. Try this: In [17]: naive_convolve_float = naive_convolve / 255 In [18]: rank_mean_float = rank_mean / 255 In [19]: plt.imshow(np.abs(naive_convolve_float - rank_mean_float), cmap='magma') Out[19]: <matplotlib.image.AxesImage at 0x10d098fd0> Result: [cid:47EBAC9317A4470AA3DEF2C02C973463] There is a smaller difference also. Internally, filters.rank uses a fancy rolling histogram algorithm with integer data values. This means that the result of the rank_mean is only approximately accurate, essentially to within integer rounding (good enough for most real-world uses), while the convolve2d code gives you an exact value (to within floating point error). Hope this helps! Juan. On 2 Nov 2017, 7:54 PM +1100, Poughon Victor <Victor.Poughon@cnes.fr>, wrote: Hello, I looks like skimage.filters.rank.mean and scipy.signal.convolve2d don't output exactly the same images. When doing: image = data.coins() K = np.ones((11, 11)) rank_mean = rank.mean(image, selem=K) naive_convolve = convolve2d(image, K, mode="same") / K.sum() All output pixel are different, with an absolute difference varying randomly between 0 and 1. Of course there's also a massive difference at the border, but that's expected because convolve2d treats image boundaries differently. But even in the center of the image all pixels are different. I've made a test script with an illustrated output image, you can check it out in this gist: https://gist.github.com/vpoughon/b4afc76ce5dc681fda9d0550d41359d3 Am I doing something wrong? Thanks, Victor Poughon _______________________________________________ scikit-image mailing list scikit-image@python.org https://mail.python.org/mailman/listinfo/scikit-image
participants (2)
-
Juan Nunez-Iglesias -
Poughon Victor