Object detection at multiple scales improves accuracy

For detecting objects in natural visual scenes, several powerful image features have been proposed which can collectively be described as spatial histograms of oriented energy. The HoG [3], HMAX C1 [12], SIFT [10], and shape context feature [2] all represent an input image using with a discrete set of bins which accumulate evidence for oriented structures over a spatial region and a range of orientations. In this work, we generalize these techniques to allow for a foveated input image, rather than a rectilinear raster in order to improve object detection accuracy. The system leverages a spectrum of image measurements, from sharp, fine-scale image sampling within a small spatial region to coarse-scale sampling of a wide field of view. In the experiments we show that features generated from the foveated input format produce detectors of greater accuracy, as measured for four object types from commonly available data-sets.