MODELING AND QUALITY EVALUATION BASED ON VISUAL PERCEPTION OF HALFTONE REPRODUCTION

Abstract

This paper introduces model of halftone image reproduction based on TVI of printing process and human visual perception. That gives opportunity to generate images for dataset and to calculate loss function without printing images after each epoch of neural networks training which can be used for creating new algorithm of digital halftoning. In this paper quality of modeled AM and FM screens is compared to offset prints on uniformity, sharpness, noise level and structural similarity. This quality metrics can be used as loss function for neural network training. The adequacy of the model presented was verified. Neural networks nowadays wildly used for image processing and they can be also used for digital halftoning algorithms in relation to printing. Dataset of images and loss function are required to train neural networks.