[Gimp-developer] Hints on image reduction algorithm

[Vincent Cadet <v_cadet yahoo fr>]

Hello people.

I am looking for an algorithm to reduce a screen image to a very small size, e.g. 3x2 pixels width x height so as each pixel is the average colour (RGB) of the corresponding screen area. For the given size 3x2, the screen would be divided into 3 columns and two rows, which means a 1920x1080 display would be comprised of 6 areas of 640x360 pixels like this:

    1   2   3
  +---+---+---+
4 | o | o | o | 5
  +---+---+---+
6 | o |   | o | 7
  +---+---+---+

Stripes 1 to 7 correspond to areas with an o sign.

I am planning to build an ambi-light system with LED stripes I bought at IKEA; 3 stripes on top and 2 stripes aside of the monitor. Each stripe would reflect the average colour of its own area on the screen, hence the algorithm.

There would be around 10-15 computation per second. The project is made of an external USB module that drives the LED stripes. It should receive only an array of values (5 RGB values, exactly). The image reduction should take place either in a screen grabbing daemon or in a modified VNC server, I haven't evaluated yet which one is best for that. In short the code will run on a Linux machine. Mine for short, it's a Gentoo 64-bit kernel 3.4 that runs on Xorg.

There are Arduino kits for the hardware part so I'm focusing on the software part right. Most algorithms I saw are screen grabbing and involve nested loops for each pixel and I'm sure there's a better way. (E.g. MMX instructions?) I thought of JPEG reduction since it corresponds to what I need if you keep only the average from the spectrum. So my question is: can the JPEG compression algorithm be effectively used and optimized to get that to work? Or if you have better ideas or hints, I'm open.

Thanks a lot in advance.