Mailman 3 June 2016 - scikit-image

Re: skimage and opencv
by Emmanuelle Gouillart Sept. 22, 2016

Sept. 22, 2016

Hi FranÃ¯Â¿Â½ois, that's an excellent question, and not a troll :-). Opencv is a very powerful library, but it focuses primarily on computer vision (feature detection and extraction, classification, ...), as opposed to image processing in general (with other tasks such as denoising, segmentation, ...). The other big difference is that skimage builds on numpy ndarrays, and uses the full power of the numpy API (including of course the basic facilities for processing arrays as images that come with numpy), as well as some of scipy functions (you could have added scipy.ndimage to your list -- a few functions in skimage are wrappers around scipy.ndimage, that exist for the sake of completeneness). One important consequence is that algorithms working for 3-d or even n-d images can be easily implemented in 3-d/n-d in skimage, whereas opencv is restricted to 2-D images (as far as I know). Thanks to the use of numpy arrays, the API of skimage is also quite pleasant for a numpy user, more than the API of opencv. A related difference is that skimage is written in python and cython, whereas opencv is a C++ library. The two libraries attract a different crowd of developers, and a Python/Cython toolkit based on numpy arrays is easier to develop and maintain inside the Scientific Python ecosystem. I'm sure that other devs/users will have things to add to this discussion! Cheers, Emmanuelle On Thu, Dec 27, 2012 at 02:06:08PM -0800, FranÃ¯Â¿Â½ois wrote: > Hi users and devs, > It came to my knowledge that another python library (based on C++ and C > codes) for image processing exists too : opencv > I understand that numpy intregrates some basic features and we need some > advanced features but I have the feeling that skimages is redoundant with > opencv in some ways. > What's the position of skimage about that? (Don't read this question as a > troll but like a real question). > I mean that similar features exist in both. Would not be possible to > reuse/integrate opencv or merge? what's the reason for keeping them apart? > My observation is there is 4 libraries to manipulate images: > * PIL > * numpy > * skimages > * opencv > That's a lot. > Cheers,

3 2

Translating Multi-Otsu Threshold from java
by Alexandre Fioravante de Siqueira Aug. 11, 2016

Aug. 11, 2016

Hi everyone, I was thinking about using multi-Otsu thresholding for some research. I am used to use the ImageJ available plugin, http://imagej.net/Multi_Otsu_Threshold. Could you give me a hand on writing such a function which could do this for python? Maybe putting it into scikit-image. Or else, could you point me a better alternative? Thank you very much! Alex

2 5

Re: Dropping support for 2.7 in 2020
by Juan Nunez-Iglesias July 4, 2016

July 4, 2016

I would drop 2.7 sooner. NumPy/SciPy dropping it is the absolute cutoff, but there’s no reason why we can’t jump the gun (and lead rather than follow). Py2.7 users can make do with older releases. As Stéfan mentioned, this is not about erasing Py2.7 support, but not releasing new features on Py2.7. Very different. I envision that we stop producing Py2.7 releases after 0.13, but we call that an “LTS” release which will get bugfix backports until 2020. imho, with Py3.4 and especially Py3.5, Py3-only suddenly became very attractive. My three favourite features are type annotations, keyword-only arguments (these two together make it much easier to produce correct code and debug), and the @ matrix multiplication operator. The latter makes linear algebra code *so* *much* nicer to read and write, and we have our fair share in scikit-image. For my own projects, I am now Py3.5-only, always. Finally, as others have mentioned, but is worth restating: - Conda allows user-space installs of Python versions and packages, so you don’t ever depend on your sysadmin-controlled environment. - It is *absolutely false* that conda packages are only useful on Windows — I have failed to compile scipy both on OSX and Ubuntu boxes. - It is also *not* a requirement to download the monolithic Anaconda distro — with miniconda (which should be the default), you just install precisely what you need. - With conda-forge we now have a community-run repository of binary conda packages. So, in short, the switch to Py3 is easier than ever for *all* users, some might just not realise it yet ;), and switching to a Py3-only programming environment is a boon to all the scikit-image devs… though again some might not realise it yet ;). Juan. On 21 May 2016 at 12:20:38 PM, Nathaniel Smith (njs(a)vorpus.org) wrote: On May 20, 2016 07:30, "Michael Sarahan" <msarahan(a)gmail.com> wrote: > I don't think this is a concern for cython or scikit-image, but many people at bumping into the language support limit in the C++11 sense with Python 2.7 on Windows. Since VS 2008 is the de-facto standard compiler for Python 2.7, people are unable to use C++11 code in modules for Python 2.7. Some people use newer compilers anyway, which sometimes works, but is mixing runtimes, and can lead to bugs or crashes. Many people would like to support Python 2.7 using a different compiler for the whole ecosystem. One example is Ilastik, by folks at HHMI, using VS 2012 to have a custom stack: https://github.com/ilastik/ilastik-build-conda I think getting mingwpy finished and polished is probably an easier solution for this problem than forking the entire py27-on-windows ecosystem :-) -n -- You received this message because you are subscribed to the Google Groups "scikit-image" group. To unsubscribe from this group and stop receiving emails from it, send an email to scikit-image+unsubscribe(a)googlegroups.com. To post to this group, send email to scikit-image(a)googlegroups.com. To view this discussion on the web, visit https://groups.google.com/d/msgid/scikit-image/CAPJVwB%3DjRQtHn9Xj0xwv3hxtD…. For more options, visit https://groups.google.com/d/optout.

3 3

Import error when importing from skimage.transform
by sja353＠nyu.edu July 1, 2016

July 1, 2016

The environment I'm using is a Jupyter notebook running python 3.5. I can import other packages, but importing anything from skimage.transform results in the following error: ImportError Traceback (most recent call last)<ipython-input-19-741b6a154664> in <module>()----> 1 from skimage.transform import resize /home/jupyter/anaconda3/envs/py35/lib/python3.5/site-packages/skimage/transform/__init__.py in <module>()----> 1 from .hough_transform import (hough_line, hough_line_peaks, 2 probabilistic_hough_line, hough_circle, 3 hough_ellipse) 4 from .radon_transform import radon, iradon, iradon_sart 5 from .finite_radon_transform import frt2, ifrt2 /home/jupyter/anaconda3/envs/py35/lib/python3.5/site-packages/skimage/transform/hough_transform.py in <module>() 2 from scipy import ndimage as ndi 3 from .. import measure----> 4 from ._hough_transform import (_hough_circle, 5 hough_ellipse as _hough_ellipse, 6 hough_line as _hough_line, ImportError: cannot import name '_hough_circle' We've already tried uninstalling skimage, updating cython and then re-installing. Anyone else encounter this problem?

3 3

Speedy IO for multi-page compressed TIFF
by Robert McLeod June 30, 2016

June 30, 2016

Hi everyone, I have some questions about IO plugins. Namely I want to import image stacks, say 20 x 4096 x 4096 x int8, that have been compressed (LZW normally). I did some benchmarking of various skimage.io plugins: 'tifffile' was about 16 s 'pil' was about 5.5 s 'freeimage' was about 0.13 s but only imports page 0 ! If we project the 'freeimage' time it's about twice as fast as PILlow, so I would prefer to use it as the preferred plugin, but the single-page issue is a show stopper. Can anyone comment on the 'freeimage' plugin, since I know in C++ it's capable of handling multipage TIFFs what the problem is here? Sincerely, Robert

1 0

Image differencing / Change detection : Implementation problem
by Jared Lewis June 26, 2016

June 26, 2016

Hi all, I am attempting to create an image differencing script using scikit-image. Essentially, I would like to subtract one image from the other. Since the images are numpy arrays it should be possible to calculate the difference between images and then plot the result as an image. However, my result is consistently an array with a singular value, i.e. values are all the same. Below is a sample script using the built-in skimage imagery data. In this case the output is an array where all values are 156. Interestingly, when I add the image objects, it works. Any help would be much appreciated. import skimage from skimage import io, data import matplotlib.pyplot as plt import numpy as np im = data.moon() im2 = data.moon() + 100 im3 = im -im2 plt.imshow(im3, interpolation= 'nearest')

2 1

Re: Genetic Algorithms based feature extraction algorithm (by minimizing a metric)
by Josh Warner June 23, 2016

June 23, 2016

Genetic algorithms are good at minimizing difficult, non-convex spaces so I imagine the foundations for such would be useful for image segmentation via expectation minimization as well. Josh On Tuesday, June 21, 2016 at 5:04:49 PM UTC-5, Juan Nunez-Iglesias wrote: > > Hi Hakim! > > I haven't seen something like what you're mentioning but I'm intrigued! If > you can point to papers etc that implement something like this, we might be > interested in including it in scikit-image. However, we only accept > established algorithms, not the latest and greatest thing. > > Juan. > > On Sun, May 29, 2016 at 6:46 PM, Hakim Benoudjit <h.benoudjit(a)gmail.com> > wrote: > >> Hello, >> >> Is there a pixel-based region growing algorithm implemented with >> Scikit-image that would be employed for feature extraction on an image? >> Basically, the kind of algorithms I'm looking for (or just checking if >> someone came across it, to see if I'm in the right direction) works by >> adding pixels to the detected feature based on the minimization of a >> certain metric. Potentially, a pixel can be removed if the metric is not >> optimized when this pixel is added. >> >> I'm thinking of implementing this feature extraction algorithm using >> Genetic Algorithms (GA), but I can't think of a way to make it possible to >> backtrack (i.e. remove a pixel). >> Anyone encountered something similar even if it's not using GAs? >> >> -- >> You received this message because you are subscribed to the Google Groups >> "scikit-image" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to scikit-image+unsubscribe(a)googlegroups.com. >> To post to this group, send email to scikit-image(a)googlegroups.com. >> To view this discussion on the web, visit >> https://groups.google.com/d/msgid/scikit-image/b151d21c-cf0b-4892-8910-f89f… >> <https://groups.google.com/d/msgid/scikit-image/b151d21c-cf0b-4892-8910-f89f…> >> . >> For more options, visit https://groups.google.com/d/optout. >> > >

1 0

Genetic Algorithms based feature extraction algorithm (by minimizing a metric)
by Hakim Benoudjit June 21, 2016

June 21, 2016

Hello, Is there a pixel-based region growing algorithm implemented with Scikit-image that would be employed for feature extraction on an image? Basically, the kind of algorithms I'm looking for (or just checking if someone came across it, to see if I'm in the right direction) works by adding pixels to the detected feature based on the minimization of a certain metric. Potentially, a pixel can be removed if the metric is not optimized when this pixel is added. I'm thinking of implementing this feature extraction algorithm using Genetic Algorithms (GA), but I can't think of a way to make it possible to backtrack (i.e. remove a pixel). Anyone encountered something similar even if it's not using GAs?

2 1

Do all functions return float64 dtype?
by Simone Codeluppi June 21, 2016

June 21, 2016

Hi I was just wondering if the majority of the skimage functions return float64 as default dtype. I am processing a big dataset and saving the data as float64 increase the size quite a bit (compared to float32). Is there also an additional computational cost in processing float64 instead of float32? Beside the precision level is there any advantage in using float64? Thanks a lot! Simone

2 1

Re: Interesting to include a the Scattering transform in sci-kit image?
by Sira Ferradans June 18, 2016

June 18, 2016

Hello, I have one implementation question now: In order to implement the scattering, we need the wavelet transform, with morlet filters. I saw that scikit-image has support for pywt, but morlet filters are not supported now. This leaves the only option of implementing the wavelet transform directly. I thought that a good idea would be to define a list of list for all the filters, that would be saved in the fourier domain. In this context the convolution of the signal with the filter would be implemented as a multiplication in the Fourier domain. Is this implementation ok? do you have any comments, maybe references where to see a similar code already available in scikit-image? Thanks a lot for the possible comments and recommendations. On Tuesday, June 14, 2016 at 11:30:03 PM UTC+2, stefanv wrote: > > Hi Sira, Michael > > This looks very interesting! The papers you mention are well cited, and > the classification results promising (I did some texture classification on > rock slices during my masters, and even then it seemed like wavelets would > come to dominate this domain). > > I look forward to your PR & gallery example. > > Thanks! > Stéfan > > On 13 June 2016 at 06:02, Sira Ferradans <sira.fe...(a)gmail.com > <javascript:>> wrote: > >> Dear Sci-kit image developers, >> >> >> at the ENS (Paris) we are planning on implementing a Python version of >> the Scattering transform 2D and 1D. The scattering transform has proven to >> be very powerful as a descriptor for image classification and signal >> analysis. We thought that it may be useful to integrate the 2D version in >> the sci-kit image package, since it aligns well with the software package >> and the community. The idea would be to implement the functionalities >> following as much as possible the APIs you already have for similar >> functions. More specifically, we will need the following (approximately): >> >> >> -Morlet wavelet: Closely related to the Gabor wavelet, >> so it should take into account its API >> -Scattering transform: output the scattering transform >> coefficients, either for display or (as a vector) for learning purposes. >> >> >> We are attaching a small tutorial (better visualized if you download it) >> that compares the performance of the first order scattering coefficients >> computed with Gabor filters, versus the coefficients you extract in the *'Gabor >> filter banks for texture classification' >> <http://scikit-image.org/docs/dev/auto_examples/features_detection/plot_gabo…>* >> tutorial. The goal of this ipython notebook is to show that the >> implementation can be easily integrated in you library while providing a >> powerful tool for image analysis. >> >> If you think this is a good idea, please let us know. Moreover, it would >> be great if you could give us some guidelines that you think would make the >> process easier. We will be adhering to the instructions given on the >> contributions page, but please don't hesitate to give feedback on our PR! >> >> >> Best Regards, >> >> Michael Eickenberg and Sira Ferradans. >> >> -- >> You received this message because you are subscribed to the Google Groups >> "scikit-image" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to scikit-image...(a)googlegroups.com <javascript:>. >> To post to this group, send email to scikit...(a)googlegroups.com >> <javascript:>. >> To view this discussion on the web, visit >> https://groups.google.com/d/msgid/scikit-image/ad063b83-f44d-463a-85fe-159b… >> <https://groups.google.com/d/msgid/scikit-image/ad063b83-f44d-463a-85fe-159b…> >> . >> For more options, visit https://groups.google.com/d/optout. >> > >

3 2