Multi-resolution or Pyramidal Analysis

I have to run some complex code on very large images, and this is taking so much computing resources, not to mention a loooong time. I'd like to speed up the process through multi-resolution analysis.

The idea is to run the computation on a low resolution image(generated using imresize, or wavelets decomposition) and then mapping detected points (coordinates) to the original (much larger) image. Let's assume I have 10 coordinates detected on the sub-sampled image.

How can I map these coordinates to the original image? Appreciate some help here.

Thanks

Answer 1

[Image Analyst]

That seems rather obvious. If you cut the size down by half, the midpoint, say, of the image will be half way to where it was in the fullsize image.

By the way, have you seen quadtree() in the Image Processing Toolbox?

Answer 2

[zephyr]

Yes, the answer is rather obvious, isn't it. I guess I asked prematurely, bcoz I arrived at the same answer myself. I used impyramid() function to cut down the image resolution in half. I guess imresize() does the same job, if the scale factor is 0.5.