Modeling the Histogram of the Halftone Image to Determine the Area Fraction of Ink Yat-Ming Wong May 8,1998 Advisor: Dr. Jonathan Arney
Background n Drawing useful information from an image is important in various fields that depend upon them n Tools used to interpret an image need to be good enough to give meaningful data
Histogram n The histogram is a tool that gives a graphical interpretation of an image n It give us an idea of the make up of the image, such as the amount of ink in its composition
Histogram n The image is read pixel by pixel for their reflectance values R 1,9 = 0.1 R 7,10 = 0.9
Histogram
Histogram of halftone dots Ink Population Paper Population
Histogram n Segmentation of the histogram has so far been done by visual approximation n Visual approximation is a highly inaccurate method of measurement in cases where data needs to be in significant figures
Threshold Threshold, R T (?)
Solution Models to segment histogram computationally: Gaussian Model Straight-Edge Model
Gaussian Model Reflectance G1G1 G2G2 G 1 +G 2
Gaussian Model +
f(i) = F*G 1 (R) + (1-F)*G 2 (R) R1R1 R2R2 11 22 F 1-F REFLECTANCE G 1 +G 2
Sum of two gaussians vs. offset lithographic print data PROBLEM REFLECTANCE G 1 +G 2 Data
Sum of two gaussians vs. inkjet “stochastic halftone” data REFLECTANCE G 1 +G 2 Data PROBLEM
Straight Edge Model Halftone dots are a collection of edges
Straight Edge Model Model of the Halftone Reflection Distribution as a Single “Equivalent Edge” H R
Model the Halftone “Equivalent Edge Vary F H R
Model the Halftone “Equivalent Edge” H Change R min or R max R
Model the Halftone “Equivalent Edge” x scan R x where:
R x The Model H R 0 1
The Noise Model R S(R) Add A Noise Metric Assume A Reflectance Variation
H R 0 1 * S(R) The Noise Model R
Straight Edge Model RminRmax F 1-F a
Straight edge model vs. offset lithographic print data H(R) R
Straight edge model vs. inkjet “stochastic halftone” data H(R) R
Comparison of models in matching offset lithographic print data Sum of two gaussiansStraight Edge vs.
Comparison of models in matching inkjet “stochastic halftone” data Sum of two gaussiansStraight Edge vs.
Automated computation n Program written in Visual Basic n Opens up a data file and automatically find the best computational match by looking for the set of variables that yields the lowest RMS deviation value.
Problems with the straight edge model H(R) R R Expand
Problems with the straight edge model H(R) R R Expand
Conclusion n Model fits well for R i and R p close to each other n For R i and R p widely spaced, a single noise metric is inadequate.
The End