An error bound of relative image blur analysis

A lower bound of the relative image blur estimation error of a pair of images is derived analytically in this paper. This error analysis result shows that the optimal camera parameters for obtaining the most accurate depth from defocusing (DFD) results are strongly related to the object depth. Therefore, DFD methods using only two images usually cannot achieve the most accurate depth estimates. Instead of using only two images, a sequence of images focused at different depths should be provided in order to obtain accurate DFD results. With some prior knowledge of the object depth, the derived error bound is used to select optimal image pairs from the input image sequence for computing DFD. Real experimental results show that the resulting DFD method outperforms both a DFF method and a DFD method using two images.