Re: [OctDev] regionprops fixes n-dimensional fixes

[Jordi G. H. <jo...@oc...>]

On 6 September 2012 04:45, Søren Hauberg <so...@ha...> wrote:
>
> On Sep 5, 2012, at 11:51 PM, Jordi Gutiérrez Hermoso wrote:
>
>> regionprops does a bunch of things, most of which make sense for
>> N-dimensional images. I attach a proposed diff to address this. Can
>> anyone give me some review before I commit it?
>
>
> Your changes seem good. A few comments (though not a complete review):

I have just committed my changes:

    http://octave.svn.sourceforge.net/viewvc/octave/trunk/octave-forge/main/image/inst/regionprops.m?r1=10977&r2=10976&pathrev=10977

I ended up doing a lot more changes than the really bad diff I showed
you originally. I did some Nd indexing tricks there, and I used
nthargout, which is a function I wrote for Octave 3.6.x. I hope this
is acceptable.

> 1) You check if you should use 'bwlabel' or 'bwlabeln'. Shouldn't
> this check rather be part of 'bwlabeln', i.e. for 2D inputs,
> 'bwlabeln' can just call 'bwlabel' ?

In Matlab, bwlabel and bwlabeln use different algorithms. I didn't
analyse the bwlabel algorithm too closely, but I suppose it's better
suited for 2d images than the bwlabeln algorithm (union-find).

> 2) Not all properties are meaningful for ND input. The previous code
> checked that the input was 2D and gave an error otherwise. You have
> removed this check, which makes sense given that we now support ND
> input. But shouldn't there be some input checking for properties
> that only work for 2D input?

I think most properties make sense for ND input. I warned and refused
to return a result for those that seemed to make no sense for ND
input. These were only perimeter and orientation.

> 3) I guess the documentation needs to be updated as well?

Oh, yes. I suppose. I'll do that later. I gotta rush to make this work
for some deadline at work. ;-)

Since this seemed like a significant overhaul, I added myself to the
copyright line.

- Jordi G. H.