I have a DB of images. My aim is to find the image with highest matching to a given image. Therefore, I would like to apply a transformation on the destination image, which will assist in comparing a small number of features.

Using FFT/DFT and a low pass filter can bring to a comparison between a minimal set of features, but will not assure a good match in observer's perspective.

Therefore, I'm looking for a transformation which considers the human eye frequency response CIE oriented.

Is somebody familiar with such transformation? Can you think of other ways to compare images using human eye response features?

Thank you