Neural Net Classification and Low Distortion Reconstruction to Halftone Images

Abstract

The objective of this thesis is to reconstruct gray-level images from halftone images. We develop high performance halftone reconstruction methods for several commonly used halftone techniques. For better reconstruction quality, image classification based on halftone techniques is placed before the reconstruction process so that the halftone reconstruction process can be fine tuned for each halftone technique. The classification is based on simplified one-dimensional correlation of halftone images and processed with neural networks. The classification method reached 100% accuracy in our experiments. For image reconstruction, we develop LMS adaptive filter (LMS) method, minimum mean square error (MMSE) method, and hybrid method. The hybrid method yields best reconstruction image quality and high processing speed. In addition, the LMS method generates optimal image masks which can then be applied to MMSE and hybrid methods to setup optimal reconstruction tables.