In an embedded C app, I have a large image that I'd like to rotate by 90 degrees. Currently I use the well-known simple algorithm to do this. However, this algorithm requires me to make another copy of the image. I'd like to avoid allocating memory for a copy, I'd rather rotate it in-place. Since the image isn't square, this is tricky. Does anyone know of a suitable algorithm?
Edited to add clarification, because people are asking:
I store an image in the usual format:
// Images are 16 bpp
struct Image {
int width;
int height;
uint16_t * data;
};
uint16_t getPixel(Image *img, int x, int y)
{
return img->data[y * img->width + x];
}
I'm hoping to move the contents of the data
array around, then swap over the width
and height
member variables. So if I start with a 9x20 pixel image, then rotate it, I'll end up with a 20x9 pixel image. This changes the stride of the image, which complicates the algorithm a lot.
This might help: In-place matrix transposition.
(You might also have to do some mirroring after the transposition, as rlbond mentions).