AFAIK image recognition already works on down-sampled images, even if the high res version is available, because the algorithms are too expensive and the details are not that important or even hurt.