This lecture focused on techniques involving image quantization, including halftoning, patterning, dithering, and the direct binary search.
Halftoning / Patterning
For one reason or another, an image may contain more color (and a larger bit depth) than is needed or can be displayed with. Therefore, the number of possible levels for a color must be reduced. Halftoning involves reducing the number of colors in an image to a smaller number. In extreme cases, an image may have to be reduced to two colors, usually black and white. Patterning may be used in this case, where the two colors must be blended in a way that it gives the appearance of a particular shade.
· http://www.cs.ualberta.ca/~juancho/paper_copies/htmldither/htmldither.html - Compares different ways of performing halftoning, with an emphasis on trying to reduce artifacts in the modified images. [3]
· http://www.ee.caltech.edu/mese/halftone/ - Contains a number of papers on halftoning techniques and algorithms. [4]
· http://www.adobe.com/support/techguides/printpublishing/scanning/psscanning01c.html - An explanation on how color images are halftoned, focusing on the CMYK color space. [3]
Dithering
In this method, the shades of a color in an image are quantized using a threshold. A dither matrix is used, which contains a number of arbitrarily placed threshold values. While there is no universally agreed upon dither matrix, the goal is to design a matrix that can prevent image artifacts (like horizontal bands), and forms a growth pattern in the image.
· http://www.cs.ualberta.ca/~juancho/courses/pub/311/dithering.html - Contains a little information on how an image can be dithered using a matrix. [2]
· http://ourworld.compuserve.com/homepages/compuphase/riemer.htm - Explanation of a dithering algorithm called the Riemersma dither. [3]
· http://www.whisqu.se/per/docs/graphics31.htm - A small message that contains a few example dither matrices for different purposes, if you’re having trouble deciding what you should put in a matrix. [2]
Direct Binary Search
The direct binary search attempts to improve image quality by adding a model of a visual system. The visual system compares the grayscale version of an image, as well as a possible halftone candidate. An error is determined between the two. The pixel in question is modified, and if the error decreases as a result, the modified pixel is retained, otherwise it is ignored.
· http://www.msoe.edu/~taylor/papers/OSA96.pdf - An explanation of a direct binary search algorithm using a hard circular dot overlap model, co-authored by Dr. Taylor himself. [4]
A rating appears after each link (in square brackets). The rating indicates the level of content provided by the link. A rating of [1] indicates very little content while a rating of [5] indicates a plethora of content is provided.