Abstract
In this paper, we present a novel computational camera, namely, gratingbased computational camera (GCC). The camera utilizes diffraction grating to generate a ghosting image, which results from the diffraction grating physical properties in optics and contains depth information for image refocusing. Since the point spread function (PSF) of the ghosting image is spatial variant, we propose a method of estimating the PSF from the ghosting image captured by GCC. Our method first estimates the PSFs by non-blind deconvolution. We then cluster the PSF map into several clusters with various depths and compute the variance of PSF in each cluster as depth features. Furthermore, principal component analysis (PCA) is applied before clustering to reduce the dimensionality of the features. Since the distortion on GCC is not avoidable, the third step before refocusing is to restore the grating-degraded image by designing a PSF for de-ghosting. Our experimental results show the efficacy of the proposed scheme.