In this thesis, we deal with the problem of converting low-resolution color images into images of a limited, predetermined, color palette. To achieve the best results, dithering is used for such problems. Dithering is a concept of intentionally adding noise to reduce the presence of artifacts and generally achieve better results.
We propose four new dithering techniques that are based on edge detection and analyze them in detail. Two of them, alongside two established dithering methods, are compared in a survey. In addition to different dithering techniques, we also compare images dithered in four different color spaces — sRGB, CIELAB, ICtCp, and Oklab.
Survey results show that the new proposed dithering techniques are not better than the existing ones in most cases. They also show that in the vast majority of cases, dithering in Oklab space is significantly better than dithering in any of the other three spaces we compared. By comparing the survey results with the outputs of DSCSI and FLIP metrics we show a high level of correlation, indicating the suitability of using these metrics for analyzing dithered images, as well as the suitability of using these metrics for analyzing images that are dithered in color spaces other than sRGB or CIELAB, which are most commonly used.
Finally, we propose several potential changes to the proposed algorithms that could bring progress in the quality of dithered images.
|