Contrast-aware Halftoning

• Published in: Computer graphics forum Vol. 29; no. 2; pp. 273 - 280
• Main Authors: Li, Hua, Mould, David
• Format: Journal Article
• Language: English
• Published: Oxford, UK Blackwell Publishing Ltd 01.05.2010
• Subjects: Algorithms; C (programming language); Computer graphics; Diffusion; Error detection; Human; I.3.3 [Computer Graphics]: Picture/Image Generation-Digitizing and scanning; Image processing systems; Perception; Preserves; Similarity
• ISSN: 0167-7055; 1467-8659
• DOI: 10.1111/j.1467-8659.2009.01596.x

This paper proposes two variants of a simple but efficient algorithm for structure‐preserving halftoning. Our algorithm extends Floyd‐Steinberg error diffusion; the goal of our extension is not only to produce good tone similarity but also to preserve structure and especially contrast, motivated by our intuition that human perception is sensitive to contrast. By enhancing contrast we attempt to preserve and enhance structure also. Our basic algorithm employs an adaptive, contrast‐aware mask. To enhance contrast, darker pixels should be more likely to be chosen as black pixels while lighter pixels should be more likely to be set as white. Therefore, when the positive error is diffused to nearby pixels in a mask, the dark pixels absorb less error and the light pixels absorb more. Conversely, negative error is distributed preferentially to dark pixels. We also propose using a mask with values that drop off steeply from the centre, intended to promote good spatial distribution. It is a very fast method whose speed mainly depends on the size of the mask. But this method suffers from distracting patterns. We then propose a variant on the basic idea which overcomes the first algorithm's shortcomings while maintaining its advantages through a priority‐aware scheme. Rather than proceeding in random or raster order, we sort the image first; each pixel is assigned a priority based on its up‐to‐date distance to black or to white, and pixels with extreme intensities are processed earlier. Since we use the same mask strategy as before, we promote good spatial distribution and high contrast. We use tone similarity, structure similarity, and contrast similarity to validate our algorithm. Comparisons with recent structure‐aware algorithms show that our method gives better results without sacrificing speed.