Optimal PSF Selection Using Second-Order Frequency Analysis for Digital Auto Focusing

Abstract

In a digital autofocusing (DAF) system, estimation or selection of an optimal point-spread-function (PSF) is the most critical factor in performance and computational complexity. The proposed PSF selection algorithm falls into the category of an a priori estimation-based method. More specifically, the algorithm first extracts a set of candidate PSFs from an a priori estimated, full set of database information using cepstrum analysis. After a part of the input image is restored using the candidate PSF to analyze the edge, the proper PSF is finally selected by comparing the differences between analyzed edges and the ideal unit-step function. While existing PSF selection methods have a common problem of excessive computational overhead when the database has a large number of PSFs, the proposed algorithm significantly reduces the amount of computation by using cepstrum analysis. As shown in the experimental results, to process a single image with a database of size 100, the proposed algorithm performs 26 times faster than the existing PSF selection method. With its enhanced implementational efficiency, the proposed algorithm is suitable for the DAF function of small, mobile imaging devices such as cellular phone cameras and compact camcorders.