Work at SourceForge, help us to make it a better place! We have an immediate need for a Support Technician in our San Francisco or Denver office.
Close
From: Richard Brown <rgbrown@gm...>  20070322 03:54:30

Hi there, I'm pretty new to python; I'm in the process of switching from Matlab. I do quite a bit of image processing in my research, and while pylab/matplotlib seems to be a great plotting library, some of the quirks seem just a little bit frustrating  I'm hoping for some enlightenment :) When I use imshow or matshow to display an array, the points with index (m, n) are displayed with the pixels centred at (0.5 + m, 0.5 + n). Is there a setting somewhere to make it so that the centres of the pixels are at the index values rather than their bottom left corners? Or must I be always adding 0.5 to things to make them appear in the right places? Thanks, I appreciate any help Richard  Richard Brown Ph.D. Candidate Dept. of Mechanical Engineering University of Canterbury, NZ 
From: Richard Brown <rgbrown@gm...>  20070322 03:54:30

Hi there, I'm pretty new to python; I'm in the process of switching from Matlab. I do quite a bit of image processing in my research, and while pylab/matplotlib seems to be a great plotting library, some of the quirks seem just a little bit frustrating  I'm hoping for some enlightenment :) When I use imshow or matshow to display an array, the points with index (m, n) are displayed with the pixels centred at (0.5 + m, 0.5 + n). Is there a setting somewhere to make it so that the centres of the pixels are at the index values rather than their bottom left corners? Or must I be always adding 0.5 to things to make them appear in the right places? Thanks, I appreciate any help Richard  Richard Brown Ph.D. Candidate Dept. of Mechanical Engineering University of Canterbury, NZ 
From: Eric Firing <efiring@ha...>  20070322 04:43:49

Richard Brown wrote: > Hi there, > > I'm pretty new to python; I'm in the process of switching from Matlab. > I do quite a bit of image processing in my research, and while > pylab/matplotlib seems to be a great plotting library, some of the > quirks seem just a little bit frustrating  I'm hoping for some > enlightenment :) > > When I use imshow or matshow to display an array, the points with > index (m, n) are displayed with the pixels centred at (0.5 + m, 0.5 + > n). Is there a setting somewhere to make it so that the centres of the > pixels are at the index values rather than their bottom left corners? > Or must I be always adding 0.5 to things to make them appear in the > right places? Coincidentally, a few days ago I made this change for matshow; I had made it in spy some time ago. I have so far left imshow alone; isn't its present behavior consistent with Matlab? That is not necessarily a good reason for leaving it the way it is, but it is reason for some caution. I suspect quite a few people may prefer it the way it is; let's see who responds, and what opinions are voiced. One way of maintaining compatibility (if it is indeed compatible now, and if this is judged desirable) would be to add a kwarg and/or rc option. But if there is a consensus that imshow should by default put index ticks at cell centers, then I will be happy to make that change. Eric > > Thanks, I appreciate any help > > Richard 
From: Richard Brown <rgbrown@gm...>  20070323 00:18:29

On 22/03/07, Eric Firing <efiring@...> wrote: > Richard Brown wrote: > > Hi there, > > > > I'm pretty new to python; I'm in the process of switching from Matlab. > > I do quite a bit of image processing in my research, and while > > pylab/matplotlib seems to be a great plotting library, some of the > > quirks seem just a little bit frustrating  I'm hoping for some > > enlightenment :) > > > > When I use imshow or matshow to display an array, the points with > > index (m, n) are displayed with the pixels centred at (0.5 + m, 0.5 + > > n). Is there a setting somewhere to make it so that the centres of the > > pixels are at the index values rather than their bottom left corners? > > Or must I be always adding 0.5 to things to make them appear in the > > right places? > > Coincidentally, a few days ago I made this change for matshow; I had > made it in spy some time ago. I have so far left imshow alone; isn't > its present behavior consistent with Matlab? That is not necessarily a > good reason for leaving it the way it is, but it is reason for some > caution. I suspect quite a few people may prefer it the way it is; > let's see who responds, and what opinions are voiced. Thanks for your timely response. Let me give you a few examples to clarify the things which I think might be relevant issues to address. (numpy and pylab imported) PRELIMINARIES # Create a 6x6 logical array with a 2x2 square near the to left xx = zeros((6, 6), dtype='Bool') xx[1:3, 1:3] = True EXAMPLE 1  imshow Trying to plot a point which should appear on the square: >>> imshow(xx, interpolation='nearest') >>> plot([2],[2], 'y.', markersize=20) The image looks correct, with the square in the top left, but the y axis is labelled backwards. Therefore when I try to plot a point in the middle of it, it misses altogether EXAMPLE 2  matshow (not your new version) >>> matshow(xx) So far so good  the y axis is the right way around >>> plot([2],[2], 'y.', markersize=20) Oops  the y axis flipped, there is a block of white at the top, and the image is now upsidedown. The point has showed up in the right place w.r.t the image though. EXAMPLE 3  off by 0.5 problem  relevant to imshow too >>> matshow(xx) Let's say I want to compute the centroid of the square blob. IMO a natural way to do this is: >>> cen = mean(where(xx), 1) >>> plot([cen[0]], [cen[1]], 'y.') This is off by 0.5 in both directions. This kind of thing is my argument for why the coordinate system should be aligned with the array indices. Matlab behaviour: In Matlab, the pixels are centred on integer coordinates corresponding to their array index. Matlab indexing is ones based, so a 2x2 image will have axes limits of 0.52.5 in each direction, with the pixel centres at (1,1), (1,2) etc. imshow in Matlab plots the array with the (1,1) coordinate in the top left, and the y axis increasing from the top down (like what matshow does here) cheers, Richard 
From: Eric Firing <efiring@ha...>  20070323 08:23:48

Richard, I have made the requested change to imshow, so it is consistent with the new matshow, and I think it makes much more sense this waybut note that it depends on your rc value for image.origin. So, whenever you get mpl from svn, or the next release (whenever that occursnot very soon, I suspect), I think you will find both matshow and imshow more to your liking. I also changed the docstring explanation for the extent kwarg, noting that it gives (left, right, bottom, top) data limits of the axes. More doc changes and other tweaks for this class of functions will be in order, when I get back to it. Eric Richard Brown wrote: > On 22/03/07, Eric Firing <efiring@...> wrote: >> Richard Brown wrote: >> > Hi there, >> > >> > I'm pretty new to python; I'm in the process of switching from Matlab. >> > I do quite a bit of image processing in my research, and while >> > pylab/matplotlib seems to be a great plotting library, some of the >> > quirks seem just a little bit frustrating  I'm hoping for some >> > enlightenment :) >> > >> > When I use imshow or matshow to display an array, the points with >> > index (m, n) are displayed with the pixels centred at (0.5 + m, 0.5 + >> > n). Is there a setting somewhere to make it so that the centres of the >> > pixels are at the index values rather than their bottom left corners? >> > Or must I be always adding 0.5 to things to make them appear in the >> > right places? >> >> Coincidentally, a few days ago I made this change for matshow; I had >> made it in spy some time ago. I have so far left imshow alone; isn't >> its present behavior consistent with Matlab? That is not necessarily a >> good reason for leaving it the way it is, but it is reason for some >> caution. I suspect quite a few people may prefer it the way it is; >> let's see who responds, and what opinions are voiced. > > Thanks for your timely response. Let me give you a few examples to > clarify the things which I think might be relevant issues to address. > (numpy and pylab imported) > > PRELIMINARIES > # Create a 6x6 logical array with a 2x2 square near the to left > xx = zeros((6, 6), dtype='Bool') > xx[1:3, 1:3] = True > > > EXAMPLE 1  imshow > Trying to plot a point which should appear on the square: >>>> imshow(xx, interpolation='nearest') >>>> plot([2],[2], 'y.', markersize=20) > > The image looks correct, with the square in the top left, but the y > axis is labelled backwards. Therefore when I try to plot a point in > the middle of it, it misses altogether > > EXAMPLE 2  matshow (not your new version) >>>> matshow(xx) > So far so good  the y axis is the right way around >>>> plot([2],[2], 'y.', markersize=20) > Oops  the y axis flipped, there is a block of white at the top, and > the image is now upsidedown. The point has showed up in the right > place w.r.t the image though. > > EXAMPLE 3  off by 0.5 problem  relevant to imshow too >>>> matshow(xx) > Let's say I want to compute the centroid of the square blob. IMO a > natural way to do this is: >>>> cen = mean(where(xx), 1) >>>> plot([cen[0]], [cen[1]], 'y.') > > This is off by 0.5 in both directions. This kind of thing is my > argument for why the coordinate system should be aligned with the > array indices. > > Matlab behaviour: > In Matlab, the pixels are centred on integer coordinates corresponding > to their array index. Matlab indexing is ones based, so a 2x2 image > will have axes limits of 0.52.5 in each direction, with the pixel > centres at (1,1), (1,2) etc. > imshow in Matlab plots the array with the (1,1) coordinate in the top > left, and the y axis increasing from the top down (like what matshow > does here) > > cheers, > > Richard 
From: Richard Brown <rgbrown@gm...>  20070326 01:05:01

On 23/03/07, Eric Firing <efiring@...> wrote: > Richard, > > I have made the requested change to imshow, so it is consistent with the > new matshow, and I think it makes much more sense this waybut note > that it depends on your rc value for image.origin. So, whenever you get > mpl from svn, or the next release (whenever that occursnot very soon, > I suspect), I think you will find both matshow and imshow more to your > liking. > Hi Eric Thanks heaps  that was exactly what I was hoping for. A small thing  I think the change might have induced a slight problem in matshow  e.g. for a 3x3 image, the pixels are in the right place, but the axis limits need to change from 0...3 to 0.5...2.5. I'm impressed with the speed of the change  seems like a very helpful community here. cheers, Richard 