Image reduction for object recognition

The problem addressed is that of the fast generation of accurate multiple downscaled copies of an input image, with an arbitrary non-integer reduction factor, for object detection and recognition. Three algorithms that can be implemented in parallel, recursive and hybrid architectures are introduced and compared in terms of their accuracy and computational complexity. It is shown that the recursive algorithm is the most advantageous in terms of computational complexity, while the parallel algorithm performs better in terms of accuracy reduction. The hybrid algorithm combines advantages of both.