Sparse projections and motion estimation in colour filter arrays

Integral image projection, computed by summing pixel rows or columns, is one of the standard tools used in computer vision algorithms for image analysis. We propose methods of reducing the complexity of image projection computation. The algorithm skips some pixels during pixel-by-pixel scanning, while preserving the resolution of original image projections. The approximated or "sparse "projections can replace the standard projections in any image processing algorithm, enabling low-cost solutions. Particularly, we consider projection-based dominant motion estimation in video and confirm the method's effectiveness by experimental results. Some variants of sparse projections can be applied directly to Colour Filter Array (CFA) images, thus allowing the algorithm to obtain high-level information on camera motion at the lowest level stage in the image processing pipeline.