Here's my situation. I'm writing a script that takes a greyscale GIF image and treats it as a 'channel' used to mix two colours and create a colourised version of it. So far things are good. It works exactly as I'd like it to, 'cept its slow. Real, real slow. Without further ado, I put together this page:

http://www.g33kz.co.uk/mh_channel_mix/source.php

You can see the whole source of the script there and test it out using the form at the bottom. I know there's a lot of work still to do, especially in error handling etc. but I want to clear up the performance issue first.

If you read the ISSUES and TODO areas of the comments you'll see a couple of ideas I've had, but before I embark upon them I wanted to run the code by a few more minds.

If I do choose to write my own gif interpreter/writer, I'll need to know how a gif is put together. I know that GIFs are a 256 colour palette followed by a bitmap, all compressed with LZW (it is LZW, right) compression. Still, I wouldn't have the first clue how to read it, decompress it and interpret it the way GD does.

Many thanks in advance. You're all beautiful.

well i don't know anything about this.

I did once use a bicubic image resizer. It actually interpolated pixels stuff and reevaluated the color. Atleast i think cause i don't understand it really. Anyhowz it is pretty fast, hope it helps out.

function imageCopyResampleBicubic($dst_img, $src_img, $dst_x, $dst_y, $src_x, $src_y, $dst_w, $dst_h, $src_w, $src_h)
{
  $scaleX = ($src_w - 1) / $dst_w;
  $scaleY = ($src_h - 1) / $dst_h;

  $scaleX2 = $scaleX / 2.0;
  $scaleY2 = $scaleY / 2.0;

  $tc = imageistruecolor($src_img);

  for ($y = $src_y; $y < $src_y + $dst_h; $y++)
  {
   $sY  = $y * $scaleY;
   $siY  = (int) $sY;
   $siY2 = (int) $sY + $scaleY2;

   for ($x = $src_x; $x < $src_x + $dst_w; $x++)
   {
     $sX  = $x * $scaleX;
     $siX  = (int) $sX;
     $siX2 = (int) $sX + $scaleX2;

     if ($tc)
     {
       $c1 = imagecolorat($src_img, $siX, $siY2);
       $c2 = imagecolorat($src_img, $siX, $siY);
       $c3 = imagecolorat($src_img, $siX2, $siY2);
       $c4 = imagecolorat($src_img, $siX2, $siY);

       $r = (($c1 + $c2 + $c3 + $c4) >> 2) & 0xFF0000;
       $g = ((($c1 & 0xFF00) + ($c2 & 0xFF00) + ($c3 & 0xFF00) + ($c4 & 0xFF00)) >> 2) & 0xFF00;
       $b = ((($c1 & 0xFF)  + ($c2 & 0xFF)  + ($c3 & 0xFF)  + ($c4 & 0xFF))  >> 2);

       imagesetpixel($dst_img, $dst_x + $x - $src_x, $dst_y + $y - $src_y, $r+$g+$b);
     }
     else
     {
       $c1 = imagecolorsforindex($src_img, imagecolorat($src_img, $siX, $siY2));
       $c2 = imagecolorsforindex($src_img, imagecolorat($src_img, $siX, $siY));
       $c3 = imagecolorsforindex($src_img, imagecolorat($src_img, $siX2, $siY2));
       $c4 = imagecolorsforindex($src_img, imagecolorat($src_img, $siX2, $siY));

       $r = ($c1['red']  + $c2['red']  + $c3['red']  + $c4['red']  ) << 14;
       $g = ($c1['green'] + $c2['green'] + $c3['green'] + $c4['green']) << 6;
       $b = ($c1['blue']  + $c2['blue']  + $c3['blue']  + $c4['blue'] ) >> 2;

       imagesetpixel($dst_img, $dst_x + $x - $src_x, $dst_y + $y - $src_y, $r+$g+$b);
     }
   }
  }
}

It uses the same kind of methodology to cycle through each pixel one at a time. Perhaps I need to implement some benchmarking code to isolate exactly which parts of the script are running slow.

Update: Great, great, great speed improvements. I noticed a huge rookie error on my part, since I was loading the channel gif each iteration (once for each pixel) which was an obvious flaw. It's down from around 8 seconds to about 0.25 seconds.

I also added channel cache which dumps the raw channel data into a file so the gif doesn't have to be parsed each time. It saves about a tenth of a second on loading times, which is nice. 0.2 is not so bad, I suppose, but I would prefer it at about 0.05.

Anyone want to suggest any other optimisations?