How to efficiently find a small logo in an image

I am trying to fine tune my image detection project and through my various tests I seem to hit a snag. If I have a logo that is small, (around 10x15), in an image my detection seems to fail and bypass it. I tried to manually crop that logo and upscale it to different resolutions, but the up-scaling causes too much noise and distorts the logo. I tried the reverse method as well by taking the logo from the image and use it to test with, thinking it should pick it up being a 1-1 comp开发者_运维百科arison, however it had reverse side effects of finding too many matches in the image.

I am using SIFT with affine transformations, and color calibration with histograms to detect my other matches. I've tried sharpening an image after up-scaling it, but that didn't work well either.

Is there anything that I could test or implement that would detect the logo? Any advice is greatly appreciated.

EDIT

Including two different sets of tests that I ran where it failed.

Nike:

Image to test logo on (Trying to get it off the collar):

Apple:

Image to test on (Cannot find enough matches to register a hit):