Blur identification from vector quantizer encoder distortion

Abstract

A method is presented for image blur identification from vector quantizer (VQ) encoder distortion. The method requires a set of training images produced by each member of a set of candidate blur functions. Each of these sets is then used to train a VQ encoder. Given an image degraded by an unknown blur function, the blur function can be identified by choosing from among the candidates the one corresponding to the VQ encoder with the lowest encoder distortion. Two training methods are investigated: the generalized Lloyd algorithm and a non-iterative discrete cosine transform (DCT)-based approach.