Fwd: possible issue in octave-forge package image (fftconv2)

[Alireza Nejati]

---------- Forwarded message ----------
From: Alireza Nejati <address@hidden>
Date: Mon, Jul 8, 2013 at 12:21 PM
Subject: Re: possible issue in octave-forge package image (fftconv2)
To: Carnë Draug <address@hidden>

Hi,

Thanks for that.

As for the other problem, it seems to be a simple matter of getting the row/column vector orientation wrong. In my version, lines 51-52 are:

    v1 = varargin{1}(:)';
    v2 = varargin{2}(:);

Changing this to:

    v1 = varargin{1}(:);
    v2 = varargin{2}(:)';

Fixes the problem for me.

Regards,

On Mon, Jul 8, 2013 at 4:46 AM, Carnë Draug <address@hidden> wrote:

On 4 July 2013 01:08, Alireza Nejati <address@hidden> wrote:
> In fftconv2.m in the image package, this is what line 47 currently looks
> like (in the latest version downloaded from sourceforge):
>
>      if ((nargin > 2) && ismatrix(varargin{3}))
>
> Since octave returns 1 for, say, ismatrix("same"), it doesn't recognize
> fftconv2(a,b,shape) properly. It thus returns errors for me. However, if I
> also add a isnumeric(varargin{3}) check, it works fine. This is what the
> corrected line looks like:
>
>     if ((nargin > 2) && ismatrix(varargin{3}) && isnumeric(varargin{3}))

Hi

that bug has already been fixed on the development version [1],
though the check we did was !ischar().

However, the function has another bug that went undiagnosed for a long
time. It returns the wrong results when you use different vectors for
the convolution in the rows and columns direction. Take a look a the
source for more details. If you could fix that one, that'd be great.

Carnë


[1] https://sourceforge.net/p/octave/image/ci/default/tree/inst/fftconv2.m