Vector Color Error Diffusion Halftoning

Mr. Niranjan Damera-Venkata
Dept. of Electrical and Computer Engineering
The University of Texas at Austin

Friday, November 5th, 4:00 PM, ENS 602

damera-v@ece.utexas.edu


Abstract

Digital halftoning is the process of conversion of continuous-tone images images (usually 8 bits/pixel) to halftone images (images with N bits/pixel, N<8). Typically for digital hardcopy applications an 8 bit/pixel image is quantized to 1 bit/pixel. Error-diffusion is a high-quality technique used to produce digital halftones. Error-diffusion is a feedback system which attempts to shape the quantization error into the high-frequency regions, where its visual impact is much smaller. I will describe a general framework for the analysis of color error-diffusion systems. I will account for correlation among the image planes (usually RGB) explicitly within the model. This model encompasses previous work in modeling greyscale error-diffusion. We will validate the model, by using it to predict important well known effects of error-diffusion such as image sharpening and noise shaping. The model also leads naturally into an interesting optimization problem for the design of the filter coefficients, for minimized visual degradation. Here the filter coefficients are matrices and not scalar valued. Also conventional convolution is replaced with matrix-vector multiply accumulates. I will show that a multifilter has an inherently parallel implementation. We show that we can reorder the computations such that the parallelism is exposed.


A list of Telecommunications and Signal Processing Seminars is available at from the ECE department Web pages under "Seminars". The Web address for the Telecommunications and Signal Processing Seminars is http://anchovy.ece.utexas.edu/seminars