| Descript |
233 p |
| Note |
Source: Dissertation Abstracts International, Volume: 58-12, Section: B, page: 6719 |
|
Supervisor: Theophano Mitsa |
|
Thesis (Ph.D.)--The University of Iowa, 1997 |
|
Halftoning has traditionally been plagued by quality tradeoffs. Generally, smoothness comes at the expense of dullness and loss of details and sharpness comes at the expense of graininess. Techniques that seek to achieve a "blue noise" distribution in the error spectrum are widely accepted as an acceptable balance of smoothness and sharpness. However, achieving a "blue noise" distribution limits image quality by allowing unconstrained high frequency error without regard to image content. The system developed in this thesis, called the Texture Model Based (TMB) halftone algorithm seeks to improve halftone image quality beyond that achieved by non adaptive blue noise techniques by adapting error distribution based on local image content |
|
The system is comprised of a multichannel model of human visual perception. The model contains a multichannel filter bank, drawn from work in texture discrimination and classification, embedded in a nonlinear processing stage. Using this model, feature vectors are defined at every pixel which characterize the perceptual features in a neighborhood about each pixel in the grayscale image. The feature vectors then serve as a guide to choosing an error constraint for every pixel that is appropriate for its neighborhood characteristics. Then an iterative halftoning algorithm is implemented to select a binary value for each grayscale pixel that minimizes the difference between the halftone and the grayscale image in a neighborhood of the pixel weighted by the error metric |
|
In order to assess the feasibility of the system, it was implemented and applied to a varied set of images. The results were assessed both qualitatively, by careful observation of specific images, and quantitatively, by an observer test. Observation indicated that the adaptive method did combine the desirable aspects of smoothness and sharpness in the same image. The results were reproductions that rendered smooth regions smoothly and texture and details sharply, while preserving overall contrast. The pleasing nature of the results was supported by the results of an observer test. The observer test indicated that the new adaptive TMB halftone method was preferred over two other non adaptive, good quality halftone reproductions, both achieving a blue noise error spectrum |
|
School code: 0096 |
|
DDC |
| Host Item |
Dissertation Abstracts International 58-12B
|
| Subject |
Engineering, Electronics and Electrical
|
|
Computer Science
|
|
0544
|
|
0984
|
| Alt Author |
The University of Iowa
|
|
