Halftone Quality Evaluation Using Color Visual Models

We conducted a monitor-based subjective study on multilevel halftone. Observers in a halftone preference experiment selected stimuli in a paired comparison design. Images were halftoned using multilevel stochastic screening and multilevel error diffusion. Stimuli were presented on a calibrated monitor from distances equivalent to viewing 300 and 600 dpi prints. We applied Thurstone's law of comparative judgments to derive an image quality scale from the paired comparison data. The derived scale values were compared to summary measures computed from variants of a color difference model (CVDM). CVDM is a multi-channel model having multiple orientation and spatial frequency mechanisms. CVDM was modified to better model the human visual system. An eye OTF front-end is added and CSF filtering and cortex transform are performed in the LAB space. We compare predictions of CVDM with S-CIELAB and observed slightly better correlation between the models and subjective scale values. On a local scale, the visible difference maps produced by the multi-channel model appear to capture worming and texturing artifacts better than S-CIELAB.