Cepstrum Analysis Based Blind Image Deconvolution
Haruka Asai, Mitsuo Eguchi · 2009
The research of image restoration that restores a latent image from a blurred image has been widely done. In this paper, we propose blind deconvolution method restoring a latent image from a blurred image degraded by a uniform two-dimensional motion blur. We focus attention on a characteristic of a cepstrum that approximation shape of a PSF appears on a cepstrum of a blurred image. Our method estimates two-dimentional PSF from a single blurred image, thus we need neither special hardware nor multiple images. First, we estimate PSF candidates from a cepstrum of a blurred image and obtain restored images by using each PSF candidate. From the restored images, we select the best restoration effect one among them as a final restored image. In order to demonstrate the effectiveness of the proposed method, we apply our method with both synthetic images and real images.