Video deblurring in complex wavelet domain using local Laplace prior for enhancement and anisotropic spatially adaptive denoising for PSF detection

This paper presents a new algorithm for video deblurring using frames before and after each scene as a multiframe observation from that scene. For this reason we develop the recently proposed algorithms that try to benefit from advantages of advanced denoising methods. At first the data is transformed to discrete complex wavelet transform (DCWT) and an initial estimate of clean data and point spread function (PSF) is obtained based on minimization of the energy criterion in gradient projection algorithm. In the next stage we improve the estimated clean data using a denoising method employing local Laplace prior and the estimated PSF is enhanced using an anisotropic spatially adaptive denoising procedure based on the local polynomial approximation (LPA) of blur operator and the intersection of confidence intervals (ICI) used for selection of window sizes of LPA. The mentioned procedure is repeated (in gradient projection algorithm) to obtain the appropriate estimations of PSF and clean data. Applying this technique for deblurring of video sequences produces better results in comparison with other methods.