Mailman 3 September 2013 - 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

Re: Contribution: circles and hough transform
by François June 18, 2014

June 18, 2014

Hi, FYI, I plan to work on the detection of ellipses quite soon. Stefan indicated to me a reference which seems fairly simple to implement. Cheers, François.

3 3

Re: import error on linux
by Payal Gupta Sept. 29, 2013

Sept. 29, 2013

hi josh... after installing cython module of skimage import on python??? me too install cython bt unfortunate skimage module not imported. On Friday, September 27, 2013 12:52:35 PM UTC+5:30, Payal Gupta wrote: > > hello.... > > i am using skimage library on raspberry pi which have linux platform but > i cant import skimage module on this. > > what do i do? > anybody know this prob. > reply >

3 2

Re: Image transformation.
by Johannes Schönberger Sept. 28, 2013

Sept. 28, 2013

> skimage.transform.warp calls scipy map_coordinates dosen't it? I believe that does not use the same interpolation method as the IDL routines. I wonder if that could be worked around though. In case of the PolynomialTransform it will always use map_coordinates, for linear transformations with a transformation matrix, it uses a much faster internal Cython implementation for order < 4. If you set order=3 it would use the same interpolation method as IDL, but you do not have the option to manually set the scaling of the sinc function. But it uses a sensible default for that.

1 0

Re: Image transformation.
by Johannes Schönberger Sept. 28, 2013

Sept. 28, 2013

Hi, You would have to do the following: 1. Setup the transformation object with the appropriate coefficients: tform = skimage.transform.PolynomialTransform(…) 2. Warp the image: img_warped = skimage.transform.warp(img, tform, order=INTERP_ORDER) INTERP_ORDER equals the Interp argument in Poly_2D of IDL. Let us know if the documentation can be improved in any way. Regards, Johannes Am 28.09.2013 um 12:40 schrieb Stuart Mumford <stuart(a)mumford.me.uk>: > Hello, > > While having a more thorough look at the transform module with respect to the image registration discussion, I thought it might be worth picking your brains on another solar based image processing issue we are having. > > When images are collected from basically all telescopes, ground or space based, a large amount of processing is required to get them to the point where you can do science on them. This normally involves dark frame or noise removal, scaling etc. The last step is to compensate for the telescope pointing, which is generally a translate, rotate and rescale operation. > > It is this last step that we are currently trying to port from the 'standard' IDL package into SunPy for the AIA instrument on NASAs SDO satellite. > > A rotate, translate and rescale transform is not hard, the problem is we really need to replicate the same algorithm as used in IDL so that the processing of the data is the same irrespective of the language used to process it. > > Without trying to show you all the gritty detail of the pipeline there is a function (for which the source code is available) called rot that is called to do the transform (here: http://www.astro.washington.edu/docs/idl/cgi-bin/getpro/library32.html?ROT) this function however calls a closed source routine called poly_2D which is described here: http://www.exelisvis.com/docs/POLY_2D.html > > To my untrained eye this looks a little like http://scikit-image.org/docs/dev/api/skimage.transform.html#polynomialtrans… ? However for polynomial transform there is no discussion about interpolation like there is for poly_2D? > > We have implemented a routine in SunPy which does a cubic affine transform in an attempt to replicate the behaviour but our testing so far is still a fair way off from the IDL implementation. > > Any advice would be as always much appreciated. > > Stuart > > > -- > 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. > For more options, visit https://groups.google.com/groups/opt_out.

2 1

Image transformation.
by Stuart Mumford Sept. 28, 2013

Sept. 28, 2013

Hello, While having a more thorough look at the transform module with respect to the image registration discussion, I thought it might be worth picking your brains on another solar based image processing issue we are having. When images are collected from basically all telescopes, ground or space based, a large amount of processing is required to get them to the point where you can do science on them. This normally involves dark frame or noise removal, scaling etc. The last step is to compensate for the telescope pointing, which is generally a translate, rotate and rescale operation. It is this last step that we are currently trying to port from the 'standard' IDL package into SunPy for the AIA instrument on NASAs SDO satellite. A rotate, translate and rescale transform is not hard, the problem is we really need to replicate the same algorithm as used in IDL so that the processing of the data is the same irrespective of the language used to process it. Without trying to show you all the gritty detail of the pipeline there is a function (for which the source code is available) called rot that is called to do the transform (here: http://www.astro.washington.edu/docs/idl/cgi-bin/getpro/library32.html?ROT) this function however calls a closed source routine called poly_2D which is described here: http://www.exelisvis.com/docs/POLY_2D.html To my untrained eye this looks a little like http://scikit-image.org/docs/dev/api/skimage.transform.html#polynomialtrans… However for polynomial transform there is no discussion about interpolation like there is for poly_2D? We have implemented a routine in SunPy which does a cubic affine transform in an attempt to replicate the behaviour but our testing so far is still a fair way off from the IDL implementation. Any advice would be as always much appreciated. Stuart

1 0

Re: import error on linux
by Payal Gupta Sept. 28, 2013

Sept. 28, 2013

python version 2.7 On Friday, September 27, 2013 12:52:35 PM UTC+5:30, Payal Gupta wrote: > > hello.... > > i am using skimage library on raspberry pi which have linux platform but > i cant import skimage module on this. > > what do i do? > anybody know this prob. > reply >

1 0

Re: import error on linux
by Payal Gupta Sept. 28, 2013

Sept. 28, 2013

hello guys... i have used 32 bit linux on raspberry pi i install cython when i import skimage then no problem but when i imoprt from skimage import io then it return that read "skimage/io/__init__.py" line 14 in module or "skimage/io/collection.py",line 12 in module like that it return i cant't understand these file lines. so what what you think where is problem? reply soon.... :( On Friday, September 27, 2013 12:52:35 PM UTC+5:30, Payal Gupta wrote: > > hello.... > > i am using skimage library on raspberry pi which have linux platform but > i cant import skimage module on this. > > what do i do? > anybody know this prob. > reply >

1 0

Re: Get border pixels of labelled region
by Payal Gupta Sept. 28, 2013

Sept. 28, 2013

hello everyone... i have problem to install skimage library on raspberry pi. how can i solve this problem.... please tell me... reply soon.... On Fri, Sep 13, 2013 at 2:19 PM, Juan Nunez-Iglesias <jni.soma(a)gmail.com>wrote: > The input is a grayscale image, not a binary image. It finds contours of a > given intensity threshold. > > The output is a list of arrays. Each array has shape (Q, 2), and it is a > set of x, y coordinates of the points in the contour. If you want to look > at the contours, you need to convert this to an image somehow. This example > uses matplotlib's `plot` function to draw the contours on the image: > > http://scikit-image.org/docs/dev/auto_examples/plot_contours.html#example-p… > > Alternately, you can round the contour coordinates to int and use those > to make a new image: > > conts = find_contours(image, threshold) > contour_image = zeros_like(image) > for c in conts: > c = np.round(c).astype(int) > coords = (c[:, 0], c[:, 1]) > contour_image[coords] = 1 > imshow(contour_image) > > > On Fri, Sep 13, 2013 at 5:26 PM, Payal Gupta <erpayal2010(a)gmail.com>wrote: > >> hi juan.... >> >> i can't understand array after using find_contour which return a ndarray. >> it is main prob i wanna find out contour array. >> >> >> On Fri, Sep 13, 2013 at 6:07 AM, Juan Nunez-Iglesias <jni.soma(a)gmail.com>wrote: >> >>> Hi Payal, >>> >>> I think we're getting somewhere. =) Thanks for the image. >>> >>> Questions: >>> 1) Are you looking to segment out different cars? That is, do you want >>> each car to have its own contour? With your current thresholding, you can >>> see that the four cars on the right are all connected, so you would get a >>> single contour of the whole blob, rather than a contour of each car. >>> 2) I just saw the doc<http://www.mathworks.com.au/help/images/ref/bwboundaries.html>for bwboundaries. Do you want the interior holes of the objects to have its >>> own contour, or do you want just a contour *around* each object and don't >>> care about the interior? Also, do you care about parent/child object >>> relationships? My guess is no to both these questions. For example, do you >>> want a contour around the windshield of each car? >>> >>> Comments: >>> >>> Note that when you do rgb2gray on a thresholded RGB image, you no longer >>> have a binary image, but a grayscale image with levels 0, 0.333, 0.667, and >>> 1.0. I think you probably mean to do rgb2gray followed by threshold. >>> >>> Either way, my suspicion is that you want to segment the cars first (get >>> one binary blob object per car), then find their contours. Am I right? >>> >>> Juan. >>> >>> >>> On Thu, Sep 12, 2013 at 8:24 PM, Payal Gupta <erpayal2010(a)gmail.com>wrote: >>> >>>> hi Juan... >>>> i have done it.... :) :) >>>> plz solve the contour problem. :( :( >>>> reply me >>>> >>>> On Thursday, September 12, 2013 9:19:36 AM UTC+5:30, Juan >>>> Nunez-Iglesias wrote: >>>> >>>>> On Thu, Sep 12, 2013 at 2:58 AM, Payal Gupta <erpay...(a)gmail.com>wrote: >>>>> >>>>>> hello... >>>>>> i m also use this prog to count white pixel and black pixel bt i cant >>>>>> save an image and cant count the pixel. >>>>>> can you help me. >>>>>> >>>>> >>>>> "white pixels" and "black pixels" will mean different things with >>>>> different images. Can you send us a sample image? >>>>> >>>> -- >>>> 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. >>>> >>>> For more options, visit https://groups.google.com/groups/opt_out. >>>> >>> >>> -- >>> You received this message because you are subscribed to a topic in the >>> Google Groups "scikit-image" group. >>> To unsubscribe from this topic, visit >>> https://groups.google.com/d/topic/scikit-image/rXRS0KT2PdU/unsubscribe. >>> To unsubscribe from this group and all of its topics, send an email to >>> scikit-image+unsubscribe(a)googlegroups.com. >>> >>> For more options, visit https://groups.google.com/groups/opt_out. >>> >> >> -- >> 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. >> For more options, visit https://groups.google.com/groups/opt_out. >> > > -- > You received this message because you are subscribed to a topic in the > Google Groups "scikit-image" group. > To unsubscribe from this topic, visit > https://groups.google.com/d/topic/scikit-image/rXRS0KT2PdU/unsubscribe. > To unsubscribe from this group and all of its topics, send an email to > scikit-image+unsubscribe(a)googlegroups.com. > For more options, visit https://groups.google.com/groups/opt_out. >

1 0

Re: Image alignment and feature extraction
by Johannes Schönberger Sept. 28, 2013

Sept. 28, 2013

Step 4 is also part of skimage. See skimage.transform.ProjectiveTransform.estimate in conjunction with the RANSAC function in skimage.measure. > Am 28.09.2013 um 00:15 schrieb "Ankit Agrawal" <aaaagrawal(a)gmail.com>: > > Hi Stuart, > >> OK I need to plead ignorance here, is there a decent reference on / can someone explain to me: >> a) what ORB is :p > ORB is a feature detection and description pipeline like SIFT and contains Oriented FAST(oFAST) and Rotated BRIEF(rBRIEF) as its feature detector and feature descriptor respectively. In short, Oriented FAST feature detector is the FAST corner detector computed along with its orientation(which is computed as in [1] Section 3.2). Rotated BRIEF is the BRIEF feature descriptor computed by rotating the test-decision pairs along the orientation of oFAST. Refer [2] and [1]. > >> b) What the difference between FAST, oFAST and rBRIEF and STAR (and FREAK?) is and which one likely to be the most appropriate for random pictures of the Sun? > Currently, I cannot answer this question in complete confidence, partly because I am yet to complete and test FREAK and partly because I am not aware of the data, or the kind of solar images that are going to be aligned. Can you give some examples of pairs of images that are required to be aligned in Solar Physics. Some queries that one would have about the data are like : Are the solar images that need to be aligned generated from the same camera but at different time intervals? Is the scaling or the zoom factor changing? Is it only the Sun's rotation about its axis that is generating these two different images or some other factor is also involved? Currently, ORB is the best bet, because of its speed and invariance to both scaling and rotation. > >> c) What else needs to be done to turn this feature detection and description work into registration? > Image Registration Pipeline can be summarized as : Detect features(keypoints) --> Compute Descriptors --> Match keypoints --> Treat one image as the reference and Estimate the Geometric Transformation required to get the second image using the matches. ORB does the first three and the 4th stage is what needs to be done. These slides[3] explain it better. > > [1] http://www.vision.cs.chubu.ac.jp/CV-R/pdf/Rublee_iccv2011.pdf > [2] http://cvlabwww.epfl.ch/~lepetit/papers/calonder_eccv10.pdf > [3] http://cs.haifa.ac.il/hagit/courses/CP/Lectures/CP05_FeaturesRegistX4.pdf > > -- > 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. > For more options, visit https://groups.google.com/groups/opt_out.

1 0