matplotlib-users

[Matplotlib-users] Blank zone with imshow()

[Dominique O. <Dom...@po...>]

When using imshow(), why does there always seem to be a blank zone along 
the southern and eastern edges of the figure? For instance:

X = rand(10,10)
imshow(X)

plots a luminance image of X, which seems fine, except for the lower and 
rightmost edges, which are blank. I may be misunderstanding the purpose 
of imshow, but skimming through the code didn't give me an answer. I am 
using matplotlib 0.52 on WinXP with either GTKAgg or TkAgg.

Thanks !
Dominique

Re: [Matplotlib-users] Blank zone with imshow()

[John H. <jdh...@ac...>]

>>>>> "Dominique" == Dominique Orban <Dom...@po...> writes:

    Dominique> When using imshow(), why does there always seem to be a
    Dominique> blank zone along the southern and eastern edges of the
    Dominique> figure? For instance:

    Dominique> X = rand(10,10) imshow(X)

    Dominique> plots a luminance image of X, which seems fine, except
    Dominique> for the lower and rightmost edges, which are blank. I
    Dominique> may be misunderstanding the purpose of imshow, but
    Dominique> skimming through the code didn't give me an answer. I
    Dominique> am using matplotlib 0.52 on WinXP with either GTKAgg or
    Dominique> TkAgg.

Hi Dominique,

Your example did point me to a small bug in the image module, but it
is mostly unrelated to what you are observing.  In the axes.py
function imshow, replace

            self.set_image_extent(0, numcols-1, 0, numrows-1) 
with 
            self.set_image_extent(0, numcols, 0, numrows) 


This only affects the tick labeling (not the actual image display)
but it was wrong before and should be changed.

Now run this script

    from matplotlib.matlab import *
    X = rand(10,10)

    subplot(211)
    im = imshow(X)
    im.set_interpolation('nearest')

    subplot(212)
    im = imshow(X)
    show()

The key thing is that the white border you are seeing arises from
interpolation.  The points on the bottom and right have no neighbors
in those directions, and so they interpolate to the background color,
which is white.

You can set the axis limits so that these regions don't appear, or use
nearest neighbor interpolation.

Let me know if these suggestions don't work for you.

JDH

Re: [Matplotlib-users] Blank zone with imshow()

[Dominique O. <Dom...@po...>]

John Hunter wrote:

>>>>>>"Dominique" == Dominique Orban <Dom...@po...> writes:
> 
> 
>     Dominique> When using imshow(), why does there always seem to be a
>     Dominique> blank zone along the southern and eastern edges of the
>     Dominique> figure? For instance:
> 
>     Dominique> X = rand(10,10) imshow(X)
> 
>     Dominique> plots a luminance image of X, which seems fine, except
>     Dominique> for the lower and rightmost edges, which are blank. I
>     Dominique> may be misunderstanding the purpose of imshow, but
>     Dominique> skimming through the code didn't give me an answer. I
>     Dominique> am using matplotlib 0.52 on WinXP with either GTKAgg or
>     Dominique> TkAgg.
> 
> Hi Dominique,
> 
> Your example did point me to a small bug in the image module, but it
> is mostly unrelated to what you are observing.  In the axes.py
> function imshow, replace
> 
>             self.set_image_extent(0, numcols-1, 0, numrows-1) 
> with 
>             self.set_image_extent(0, numcols, 0, numrows) 
> 
> 
> This only affects the tick labeling (not the actual image display)
> but it was wrong before and should be changed.
> 
> Now run this script
> 
>     from matplotlib.matlab import *
>     X = rand(10,10)
> 
>     subplot(211)
>     im = imshow(X)
>     im.set_interpolation('nearest')
> 
>     subplot(212)
>     im = imshow(X)
>     show()
> 
> The key thing is that the white border you are seeing arises from
> interpolation.  The points on the bottom and right have no neighbors
> in those directions, and so they interpolate to the background color,
> which is white.
> 
> You can set the axis limits so that these regions don't appear, or use
> nearest neighbor interpolation.
> 
> Let me know if these suggestions don't work for you.
> 
> JDH

John,

Many thanks for your reply and suggestions. I see what is happening. 
Nearest neighbor interpolation has the colors right, but i was trying to 
get "more interpolation". I guess looking at the great pictures that 
imshow() produces i was hoping for a result such as that of

X = rand(10)
pcolor( X )
shading 'interp'

in Matlab. The Matplotlib picture is just as good really, except for the 
border. Why are the other borders not white as well? Is the 
interpolation "directional"? Why aren't pixels on the border only 
interpolated with their neighbors inside the image, and not those 
outside (these have less neighbors than pixels in the middle)?

Perhaps you can point me to the part of the code (c++ i assume) which 
does the interpolation? Should i grab the CVS repository for that? Then 
maybe i can play around and see if i can achieve the effect i am looking 
for.

Thanks again !
Dominique

Re: [Matplotlib-users] Blank zone with imshow()

[John H. <jdh...@ac...>]

>>>>> "Dominique" == Dominique Orban <Dom...@po...> writes:

    Dominique> Many thanks for your reply and suggestions. I see what
    Dominique> is happening. Nearest neighbor interpolation has the
    Dominique> colors right, but i was trying to get "more
    Dominique> interpolation". I guess looking at the great pictures
    Dominique> that imshow() produces i was hoping for a result such
    Dominique> as that of

    Dominique> X = rand(10) pcolor( X ) shading 'interp'

As far as I recall, matlab's pcolor also loses an edge due to
interpolation.  Perhaps the main difference is that in matlab, the
axes limits are set by default so that you don't see it.  I remember
being surprised by this many moons ago the first time I used pcolor in
matlab.

    Dominique> in Matlab. The Matplotlib picture is just as good
    Dominique> really, except for the border. Why are the other
    Dominique> borders not white as well? Is the interpolation
    Dominique> "directional"? Why aren't pixels on the border only
    Dominique> interpolated with their neighbors inside the image, and
    Dominique> not those outside (these have less neighbors than
    Dominique> pixels in the middle)?

I'll have to think about this some more.  

There is also something funny about how the tick labeling currently
works for images, because the X[0,0] coord is upper left but is
labeled as 0,10.  Perhaps ticks should be off be default, or labeled
with the y axis descending.  Those who have opinions please weigh in.

BTW, the developer of agg, Maxim, is fairly responsive, so if you want
to pursue this issue after reading some of the code I point to below
on the agg mailing list

  http://lists.sourceforge.net/lists/listinfo/vector-agg-general

Maxim can probably provide some additional guidance.


    Dominique> Perhaps you can point me to the part of the code (c++ i
    Dominique> assume) which does the interpolation? Should i grab the
    Dominique> CVS repository for that?  Then maybe i can play around
    Dominique> and see if i can achieve the effect i am looking for.

The code is available in the matplotlib src distribution.  The module
is src/_image.cpp, which uses agg for image manipulation; see the
function Image_resize.  All the agg code is also in the matplotlib src
distro, eg, agg2/include.  agg2 doesn't have a lot of documentation -
hence I spend a lot of time reading src files, eg

    agg_conv_transform.h
    agg_span_image_filter_rgb24.h
    agg_span_image_filter_rgba32.h
    agg_span_interpolator_linear.h

The latest agg snapshot it http://www.antigrain.com/agg2.tar.gz which
has lots of examples in the examples dir.  Agg can do a lot with
images, some of which would be nice to add to the matplotlib
interface....

JDH