Estimating error diffusion kernel from error diffused images

Abstract

The problem of estimating the error diffusion kernel from error diffused images is considered. We first suggest a method for estimating an error diffusion kernel using a gray scale image and its error diffused version. The task is cast as a system identification problem and is solved using techniques from adaptive signal processing. Specifically, we define an error criterion between the error diffusion system with the true but unknown kernel, and one with an estimate of the true kernel. The estimate is then adjusted using a gradient descend type algorithm so that the error criterion is minimized. This algorithm is then combined with a projection algorithm for inverse halftoning to iteratively estimate the kernel from only an error diffused halftone.<>