A critical analysis of dithering algorithms for image processing

The authors present a critical comparison of three dithering algorithms, ordered dither, error diffusion and dot diffusion, that are based on the multichannel model of the human visual system. The techniques are analyzed and compared quantitatively with particular emphasis on finding a suitable criterion to measure the performance. It is found that all the algorithms well preserve the power spectra of the original images in the low-frequency regions. The ordered dither and dot diffusion methods generate grid patterns in the frequency domain leading to undesirable texture and moire patterns in the spatial domain. In contrast, grid patterns are absent in the error diffusion method. This perhaps explains why the error diffusion algorithm in general produces better dithered images than the others. This analysis approach provides a useful framework for a better understanding and further development of dithering techniques. >