[R-sig-Geo] spatial analyses from picture

[Barry Rowlingson]

Sometimes its easier and cheaper to farm this out to something like
Amazon's Mechanical Turk system. Basically pay people $0.01 to look at
a picture and click if its a zebra or a horse - or whatever! I suspect
you can do multiple tests on each image to make sure people don't just
randomly click.

Of course automating these analyses is a nice research project, but
sometimes you just want to get things done and cheap labour is (sadly)
cheap. In this case, cheaper than research students.

Barry


On Mon, May 19, 2014 at 2:46 PM, Forrest Stevens <forrest at ufl.edu> wrote:
> Hi Julien, you've posed a really interesting problem, and the bulk of which
> I'd argue might not be best solved using R but I'll be very interested to
> hear other people's opinions on it. A large portion of your project is
> going to rely on good image processing, so the raster and jpeg packages
> will almost certainly come in handy. R has packages that tie into external
> utilities that have better analytical capabilities than many native R
> packages provide, however.  A good example of this is ImageJ and the R glue
> package RImageJ. You'll likely need to develop a workflow that does what
> you want using one of these types of external packages in terms of image
> pre-/processing and then end up automating it in R to some degree.
>
> Though I'm more of a remote sensing specialist I've found some of the best
> image analysis packages in R coming from the medical research community.
> The next place I'd point you is to the Medical Imaging task view,as much of
> what you're trying to do might be accomplished using packages there:
>
> http://cran.r-project.org/web/views/MedicalImaging.html
>
> For example, to identify patches of color, and likely do the major task of
> separating "animal" from "background" you will likely want to use a process
> called segmentation.  The dpmixsim package is one R package I found that
> does this using modern techniques, though it looks like it might be
> stagnant. There may be others I'm unaware of. I'm afraid I rely on other
> commercial tools (ENVI/IDL and eCognition for segmentation) so I might not
> be of much use there.
>
> This blog post is a much more basic approach to image segmentation that
> might get you on the road to a solution using only the jpeg package:
>
> http://alandgraf.blogspot.com/2012/02/unsupervised-image-segmentation-with.html
>
> Anyway, the rest of your stated research problem in terms of automating a
> dorsal vs. ventral orientation can be solved in a range of ways from the
> simple (assuming you have pre-processed the image to a horizontal
> orientation, identify "animal" pixels from "background" pixels, identify
> centroid location within animal pixels/outline polygon using the rgeos
> package, draw a horizontal line through it, etc.) to something much more
> complex.
>
> Anyway, hopefully that's food for thought.
>
> Sincerely,
> Forrest
>
> --
> Forrest R. Stevens
> Ph.D. Candidate, QSE3 IGERT Fellow
> Department of Geography
> Land Use and Environmental Change Institute
> University of Florida
> www.clas.ufl.edu/users/forrest
>
>
> On Mon, May 19, 2014 at 9:03 AM, Julien Renoult <jurenoult at gmail.com> wrote:
>
>> Dear R-user
>> I have a series of spatial analyses to run with R, but I have no idea of
>> which packages to use. Actually, these are not spatial analyses sensu
>> stricto, but the principle should be the same.
>> I would like to analyse animal colour patterns from pictures. I have
>> pictures of many species, and I would like to test wether colour A is more
>> associated (ie spatially close to)  with colour B than expected by chance.
>> I thus need to :
>> - open my pictures in R (currently in jpeg)
>> - delineate body contours, colour patches A and colour patches B
>> - extract coordinates for the three shapes
>> - perform some spatial statistics
>> Also, I would like to test wether colour A is located more on the dorsal
>> than on the ventral side. I thus need to define a line (mid-body) and
>> calculate the area of colour A above and below this line.
>>
>> Please, could anyone indicate me which package(s) to use for these tasks? I
>> read the CRAN task view page for spatial data, but there are so many
>> packages !!
>> In advance, thank you very much.
>> Best regards, Julien
>>
>>
>> *Julien Renoult*
>> *Chargé de Recherche CNRS*
>> *Ph.D Evolutionary Biology - Dr. Vet. Med.*
>>
>> *UMR8218*
>> *Art Création Théorie & Esthétique*
>> *49 rue des Bergers*
>> *75015 Paris*
>> *+33615265240*
>>
>>         [[alternative HTML version deleted]]
>>
>>
>> _______________________________________________
>> R-sig-Geo mailing list
>> R-sig-Geo at r-project.org
>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>
>>
>
>         [[alternative HTML version deleted]]
>