Computing the dense Approximate Nearest-Neighbour Field (ANNF) between a pair of images has become a major problem which is being tackled by the image processing community in the recent years. Two important papers viz. PatchMatch [3] and CSH [11] have been developed over the past few years based on the coherency between images, but one major problem both these papers have is that image patches are treated as high dimensional vector features. In this paper we present a novel idea to reduce the dimensions of a p-by-p patch of color image to a set of low level features. This reduced dimension feature vector is used to compute the ANNF. Using these features we show that instead of dealing with image patches as p2 dimensional vectors, dealing with them in a lower dimension gives a much better approximation for the nearest-neighbour field as compared to the state of the art. We further present a modification which improves the ANNF to give more accurate color information and show that using our improved algorithm we do not need a pair of related images to compute the ANNF like in other algorithms, i.e. we can generate the ANNF for all the images using unrelated image pairs or even from a universal source image.