Spectral Prediction of Juxtaposed Halftones Relying on the Two-by-Two Dot Centering Model

Vahid Babaei, Roger D. HerschJournal of Imaging Science and Technology 2013 · DOI

TL;DR

This paper tackles the challenge of accurately predicting the color of printed images when using "juxtaposed halftoning," a technique where ink dots are placed side-by-side rather than overlapping. The authors demonstrate that the established Yule-Nielsen spectral Neugebauer model struggles with this new method, while the "two-by-two dot centering" model, adapted and optimized, provides significantly more accurate color predictions.

What Problem Does This Paper Address?

High-quality color reproduction in printing relies on accurately predicting the final color of a printed image based on the inks and their arrangement. Traditional halftoning methods involve overlapping ink dots, where models like the Yule-Nielsen spectral Neugebauer (YNSN) model work well by accounting for these overlaps and their resulting colors (Neugebauer primaries). However, a newer technique called "juxtaposed halftoning" places ink dots side-by-side. This is crucial for special inks like metallics or opaques. The problem is that existing YNSN models, which assume ink superposition, fail to accurately predict the colors produced by juxtaposed halftoning because the effective surface coverages of the colorants and their potential overlaps are unknown and behave differently. This paper aims to bridge this gap by adapting and evaluating a more suitable spectral prediction model for this emerging halftoning technique.

Methods

The core of this paper is the adaptation and evaluation of the "two-by-two dot centering" spectral prediction model for juxtaposed halftoning. This model, originally proposed by S. G. Wang, assumes that any halftone can be represented by a limited set of 2x2 pixel patterns.

Key methodological contributions include:

  • Adaptation for Juxtaposed Halftoning: The authors apply the two-by-two dot centering model to discrete line juxtaposed halftones, a specific implementation of this technique.
  • Predictive Calibration: To address the model's requirement for a large calibration set, they employ an estimation technique. This method predicts the reflectances of approximately 90% of the necessary 2x2 calibration patterns by measuring only about 10%, significantly reducing calibration effort. This involves establishing a linear relationship between colorant surface coverages and absorptances in a transformed space.
  • Comparison with YNSN: The authors conduct a rigorous comparison of their adapted two-by-two model against various YNSN model variants (nominal, IIS-YNSN, SDIS-YNSN) using standard and custom ink sets. This comparison is performed using "Demichel compatible" surface coverages and also with non-Demichel surface coverages to simulate more realistic scenarios.

Datasets

The paper does not explicitly introduce new benchmark datasets. Instead, it utilizes:

  • Printed Test Patches: 125 juxtaposed halftone test patches were printed using a Canon Pixma Pro9500 inkjet printer on Canon MP-101 paper. These patches were generated with varying colorant surface coverages, including those adhering to the Demichel equations and those that do not.
  • Calibration Sets: The calibration sets for the models varied in size, from 8 (for nominal YNSN) to 1072 (for full two-by-two model), with a "predictive" two-by-two model using a subset of 100 measured patterns.

Key Insights and Ideas

  • YNSN's Inadequacy for Juxtaposed Halftoning: The fundamental insight is that models designed for overlapping inks (like YNSN) fundamentally fail to predict the colors of juxtaposed halftones because the underlying assumptions about ink superposition and effective surface coverages are violated.
  • Two-by-Two Dot Centering's Suitability: The two-by-two dot centering model, by its nature of analyzing local 2x2 patterns, is inherently better suited to capture the spatial arrangement of juxtaposed dots, regardless of whether they overlap or not.
  • Efficient Calibration is Key: The paper demonstrates that a significant reduction in calibration effort (measuring <10% of patterns) is achievable without substantial loss in prediction accuracy for the two-by-two model, making it more practical.
  • Optical Dot Gain is Handled Differently: The authors highlight that the YNSN model's 'n-value' explicitly accounts for optical dot gain. In contrast, the two-by-two model implicitly handles this through its analysis of local patterns, with its own 'n-value' primarily addressing variations in ink dot thickness.

Key Results

  • Superior Accuracy of Two-by-Two Model: Across various test conditions (Demichel-compliant and non-Demichel surface coverages, standard and custom inks), the two-by-two dot centering model consistently outperformed YNSN variants in terms of prediction accuracy, as measured by ΔE94.
  • For Demichel-compliant halftones with T=7, the "Predictive 2-by-2" model achieved a mean ΔE94 of 1.39, compared to 2.32 for Nominal YNSN. The "Full 2-by-2" model achieved an even lower mean ΔE94 of 0.70.
  • Similar improvements were observed for T=11, with the "Predictive 2-by-2" model achieving a mean ΔE94 of 1.34, significantly better than Nominal YNSN's 1.84.
  • Predictive Model Performance: The "Predictive 2-by-2" model, which uses a subset of measured patterns, offered a strong balance between accuracy and calibration effort, often achieving results close to the "Full 2-by-2" model.
  • Non-Demichel Cases: The two-by-two model showed increased accuracy and robustness when dealing with non-Demichel surface coverages, where YNSN models struggled significantly.

Limitations and Future Work

  • Limitations:
    • The paper focuses on discrete line juxtaposed halftoning. The applicability to other forms of juxtaposed halftoning (e.g., clustered dot juxtaposed halftoning) might require further investigation.
  • While the predictive calibration is effective, the paper doesn't explore the absolute minimum number of measurements required for acceptable accuracy.
  • Future Work:
    • The authors express interest in verifying the prediction accuracy of their model for special inks like metallic and luminescent inks, which are primary drivers for using juxtaposed halftoning.
  • Further exploration into optimizing the predictive calibration technique for even greater efficiency could be beneficial.

Summary updated

Abstract

In a color reproduction workflow, spectral prediction models are useful for establishing the correspondence between colorant surface coverages and resulting printed halftone color. Spectral prediction models enable calculation of the color gamut and establishment of the color separation tables. Discrete line juxtaposed halftoning, a recently proposed algorithm, is characterized by the fact that colorants formed by inks and ink superpositions are placed side by side. Juxtaposed halftoning is necessary when printing with special inks such as opaque or metallic inks. In order to predict the color of classical halftones, the Yule&#x2010;Nielsen modified spectral Neugebauer model is generally used. However, this model may not predict the color of juxtaposed halftones, since the effective surface coverages of colorants and of possible colorant overlaps are unknown. In contrast, the two-by-two dot centering spectral prediction model developed by S. G. Wang enables the reflectance of slightly overlapping colorants to be captured and is therefore appropriate for predicting the color of juxtaposed halftones. Since this model requires a large calibration set, the authors use an estimation technique which predicts more than 90% of the two-by-two calibration pattern reflectances by measuring less than 10% of them. For juxtaposed halftoning, the two-by-two dot centering model offers high prediction accuracies and outperforms the different variants of the Yule&#x2010;Nielsen spectral Neugebauer model for comparable setups.
BibTeX
@article{babaei2013spectral,
  author = {Vahid Babaei and Roger D. Hersch},
  title = {{Spectral Prediction of Juxtaposed Halftones Relying on the Two-by-Two Dot Centering Model}},
  journal = {Journal of Imaging Science and Technology},
  year = {2013},
  doi = {10.2352/j.imagingsci.technol.2013.57.4.040501},
}

Loading related content...

Figure