[Digikam-devel] dcraw -h (was Re: extragear/graphics/digikam/utilities/imageeditor/editor)

[Gerhard Kulzer]

Am Mittwoch, 14. Dezember 2005 16.02 schrieb Thorsten Schnebeck:
> > > > What the problem with -h option ? The Raw decoding speed is
> > > > decreased without this option ? Please give me a feedback
> > > > before to change...
> > > >
> > > > Gilles
> > >
> > > Gilles, without the -h option it works as it should, I get
> > > the correct picture size. Reading the dcraw help it is clear
> > > that -h divides the image by 4. So there is no problem with
> > > dcraw, we have to leave out -h.
> >
> > I'm not agree with you. Why -h option working fine with other
> > file formats, and with any CR2 file that you have provide ? I
> > suspect a problem in dcraw with CR2 and -h option...
> >
> > Gilles
>
> "-h" comandline option of dcraw correspondends to "half_size=1".
> In dcraw there is this code:
>
>     shrink = half_size && filters;
>     iheight = (height + shrink) >> shrink;
>     iwidth  = (width  + shrink) >> shrink;
>     image = calloc (iheight*iwidth*sizeof *image + meta_length,
> 1);
>     merror (image, "main()");
>     meta_data = (char *) (image + iheight*iwidth);
>
> So, based on "filters" it seems to be possible, that one format
> shrinks and another does not shrink. I tested cr2 and x3f (BTW
> the latter is not part of digiKams default supported raw format
> extentions nor is any raw format part of the "Image Import..."
> filter dialog):
>
> x3f results to a full-sized image, cr2 to a half- (better
> quad-)sized image when using "dcraw -h"
>
> "filters" set or unset seems to be a result of
> int CLASS identify (int will_decode)
>
> I will forward this mail to the CRW-Decode ML and maybe
> someone can help and explain, when -h shrinks, when it does not
> shrink and what is this option all about. Can "-h" be used to
> load and decode a RAW format in best quality?
>
> Bye
>
>   Thorsten
>
> P.S.
> cr2: Canon RAW
> x3f: Foveon RAW

ufraw decodes cr2 to the correct size. Looking into its dcraw_api.c if found 
this:

     * If scale is odd we need to "unshrink" it using the info in
     * hh->fourColorFilters before scaling it. */
    if (hh->filters!=0 && scale%2==1) {
        /* I'm skiping the last row/column if it is not a full row/column */
    f->height = h = hh->height / scale;
    f->width = w = hh->width / scale;
    fujiWidth = hh->fuji_width / scale;
    f->image = g_new0(dcraw_image_type, h * w);
    f4 = hh->fourColorFilters;
    for(r=0; r<h; r++) {
        for(c=0; c<w; c++) {
        for (cl=0; cl<hh->raw.colors; cl++) sum[cl] = count[cl] = 0;
        for (ri=0; ri<scale; ri++)
            for (ci=0; ci<scale; ci++) {
            sum[FC(f4, r*scale+ri, c*scale+ci)] +=
                hh->raw.image
                [(r*scale+ri)/2*hh->raw.width+(c*scale+ci)/2]
                [FC(f4, r*scale+ri, c*scale+ci)];
            count[FC(f4, r*scale+ri, c*scale+ci)]++;
            }
        for (cl=0; cl<hh->raw.colors; cl++)
            f->image[r*w+c][cl] =
                MAX(sum[cl]/count[cl] - hh->black,0);
        if (recombine) f->image[r*w+c][1] =
            (f->image[r*w+c][1] + f->image[r*w+c][3])>>1;
        }
    }
    } else {
    if (hh->filters!=0) scale /= 2;
        /* I'm skiping the last row/column if it is not a full row/column */
    f->height = h = hh->raw.height / scale;
    f->width = w = hh->raw.width / scale;
    fujiWidth = ( (hh->fuji_width+hh->shrink) >> hh->shrink ) / scale;
    f->image = g_new0(dcraw_image_type, h * w);
    norm = scale * scale;
    for(r=0; r<h; r++) {
        for(c=0; c<w; c++) {
        for (cl=0; cl<hh->raw.colors; cl++) {
            for (ri=0, s=0; ri<scale; ri++)
            for (ci=0; ci<scale; ci++)
                s += hh->raw.image
                [(r*scale+ri)*hh->raw.width+c*scale+ci][cl];
                    f->image[r*w+c][cl] = MAX(s/norm - hh->black,0);
        }
        if (recombine) f->image[r*w+c][1] =
            (f->image[r*w+c][1] + f->image[r*w+c][3])>>1;
        }
    }
    }

Looks like that what is needed after dcraw call.

Gerhard