Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering

First Advisor

Jan P. Allebach

Committee Chair

Jan P. Allebach

Committee Member 1

Mary L. Comer

Committee Member 2

Amy R. Reibman

Committee Member 3

Robert Ulichney


This dissertation studies image quality problems associated with rendering images in devices like printing or displaying. It mainly includes two parts: clustered-dot periodic halftone screen design, and color table compression.

Screening is a widely used halftoning method. As a consequence of the lower resolution of digital presses and printers, the number of printer-addressable dots or holes in each microcell may be too few to provide the requisite number of tone lev- els between paper white and full colorant coverage. To address this limitation, the microcells can be grouped into supercells. The challenge is then to determine the desired supercell shape and the order in which dots are added to the microcell. Using DBS to determine this order results in a very homogeneous halftone pattern. To simplify the design and implementation of supercell halftone screens, it is common to repeat the supercell to yield a periodically repeating rectangular block called the basic screen block (BSB). While applying DBS to design a dot-cluster growth order- ing for the entire BSB is simpler to implement than is the application of DBS to the single non-rectangular supercell, it is computationally very inefficient. To achieve a more efficient way to apply DBS to determine the microcell sequence, we describe a procedure for design of high-quality regular screens using the non-rectangular super- cell. A novel concept the Elementary Periodicity Set is proposed to characterize how a supercell is developed. After a supercell is set, we use DBS to determine the micro-cell sequence within the supercell. We derive the DBS equations for this situation, and show that it is more efficient to implement.

Then, we mainly focus on the regular and irregular screen design. With digital printing systems, the achievable screen angles and frequencies are limited by the finite addressability of the marking engine. In order for such screens to generate dot clusters in which each cluster is identical, the elements of the periodicity matrix must be integer-valued, when expressed in units of printer-addressable pixels. Good approximation of the screen sets result in better printing quality. So to achieve a better approximation to the screen sets used for commercial offset printing, irregular screens can be used. With an irregular screen, the elements of the periodicity matrix are rational numbers. In this section, first we propose a procedure to generate regular screens starting from midtone level. And then we describe a procedure for design of high-quality irregular screens based on the regular screen design method. We then propose an algorithm to determine how to add dots from midtone to shadow and how to remove dots from midtone to highlight. We present experimental results illustrating the quality of the halftones resulting from our design procedure by comparing images halftoned with irregular screens using our approach and a template-based approach. We also present the evaluation of the smoothness and improvement of the proposed methods.

In the next part, we study another quality problem: ICC profile color table compression. ICC profiles are widely used to provide transformations between different color spaces in different devices. The color look-up tables (CLUTs) in the profiles will increase the file sizes when embedded in color documents. In this chapter, we discuss compression methods that decrease the storage cost of the CLUTs. For DCT compression method, a compressed color table includes quantized DCT coefficients for the color table, the additional nodes with large color difference, and the coefficients bit assignment table. For wavelet-based compression method, a compressed color table includes output of the wavelet encoding method, and the additional nodes with large color difference. These methods support lossy table compression to minimize the network traffic and delay, and also achieves relatively small maximum color difference.