MicroUniformity: A versatile image quality metric Bimal Mishra and - - PowerPoint PPT Presentation

MicroUniformity: A versatile image quality metric

Bimal Mishra and Rene Rasmussen Xerox Wilson Center for Research & Technology Slides presentation at IS&T’s PICS 2000 Conference

Acknowledgements

• Xerox and Fuji-Xerox image quality

standards team

– E. Dalal, K. Natale-Hoffman, P. Crean – F. Nakaya, M. Sato

• Product program people and other

“beta-testers”

– S. Reczek, S. Kuek, M. Rabbani, S. Zoltner

Lines Text Color rendition Micro uniformity Macro uniformity Tone levels Pictorial sharpness Adjacency Gloss

Micro Uniformity

DAC Color Printer IQ Attributes

See E. Dalal et al., PICS-98 Key contributors:

– Seemingly random variations

• Process noise
• Stochastic screens, etc.

– Periodic variations

• Amplitude modulated

screens

• High-frequency moire

Development of an instrumented metric

• Technology comparison; competitive benchmarking

– Current evaluation is conducted visually by expert panel – Automate evaluation of DAC Micro Uniformity – Did not have metric suitable for all technologies

• Engineering tool to assess, independently:

– Appearance of process noise (graininess) – Appearance of halftone patterns – Appearance of moire patterns – and without being affected by

• streaks
• bands

– … … ...

Description of the metric

Traditional graininess metric

• Analysis of photopic density fluctuations recorded with a long

(2-3mm) and narrow (10-20um) scanning slit.

• Amplitude spectrum modulated with human Contrast

Sensitivity Function [CSF], and integrated over frequencies.

• Density factor exp(-1.8D) applied.
• Issue: Halftone screen can dominate the measure

Filtering out the halftone frequency

• Aperture filtered halftone graininess

– Slit width set to filter out the halftone frequency; M. Maltz et

• al. 1993
• Processing of the 1D amplitude spectrum

– T. Bouk and N. Burningham 1992

• Not feasible in cases of process colors with multiple

rotated screens

Micro uniformity metric

• Image capture

– Drum scanner at 600dpi, RGB – Calibration from RGB to photopic reflectance

• Image processing

– Separation into three components – Visual filter

• Evaluation of uniformity

– Independently for the three components

• End up with 3 numbers

Image processing

Photopic R FFT Kernel

• peration

Visually filtered image Amplitude spectrum Phase spectrum Classification Binary mask structured/unstructured FFT Separate, inverse FFT, crop Structured image Unstructured image Filter HF unstructured LF unstructured

Real space Fourier space

Frequency band from f to f + Δf Average amplitude: <A>(f) Standard deviation of amplitude: S(f)

Classification of amplitude spectrum

• Statistical analysis of

narrow frequency band classifies each pixel within the band

– A(m,n) compared to <A>(f) + 3.3*S(f)

• Very-low frequency

(<0.1c/mm) region classified as unstructured f x f y

Most simple approach without f-dependency does not work (e.g., 1/f)

Single-number evaluation of each

• f the 3 image components
• Options

– Amplitude spectrum integration

• Suitable when phases are uncorrelated

– Space-domain integration

• Suitable also for localized defects (strong

phase correlation)

…evaluation of image components

• Space-domain integration

– Boundary region may contain artifacts and is cropped away – RMS of deviation from the mean – Apply lightness factor scaling

• Outputs from the analysis:

– “Visual Structure” [VS]

• Screen, high frequency moire, isolated streaks, periodic bands

– “Visual Noise High Frequency ” [VNHF]

• Process noise, stochastic screens, “graininess”

– “Visual Noise Low Frequency ” [VNLF]

• Mottle

Examples

Print samples

20mm

Plus substrate and mode variations

Combinations of 50% C,M,Y,K

Diversity of “looks”

Examples

1. Multiple rotated screens 2. Low frequency cluster dot screen 3. Frequency modulated screen 4. Rotated screen with strong rosette pattern 5. Streaks 6. Moire pattern Next

121-8

Multiple rotated screens

VisualNoise HF Visual Structure

Multiple rotated screens, 121-8

Multiple rotated screens, 121-8

VisualNoise HF contrast enhanced VNHF value

Multiple rotated screens, 121-8

Visual Structure contrast enhanced Examples VS value

Cluster dot screen, low-frequency

VisualNoise HF Visual Structure

Cluster dot screen, low-frequency

VisualNoise HF contrast enhanced

Cluster dot screen, low-frequency

Visual Structure contrast enhanced Examples

Cluster dot screen, low-frequency

Frequency modulation screen

VisualNoise HF contrast enhanced

Frequency modulation screen

Visual Structure contrast enhanced Examples

Frequency modulation screen

Rotated screens, strong rosette pattern; 70-6

Rotated screens, strong rosette pattern; 70-6

VisualNoise HF contrast enhanced

Rotated screens, strong rosette pattern; 70-6

Visual Structure contrast enhanced Examples

Removal of streaks from graininess measurement

Examples

Unprocessed image Unstructured Structured (Visual filter NOT applied here)

Moire pattern

Visual Structure

Coated paper, stronger moire pattern Uncoated paper

… moire pattern

Examples

Unprocessed image Structured image Works for 1D structure at an angle

Preliminary correlation to visual evaluations

Inspection of separated images

• Visual inspection (not IQ assessment)

– Good qualitative agreement between the separated images and visual assessment of print samples

• Weak artifacts can be seen in the “structured image” in cases

with little or no visible screen, but always lead to VS < 1 (not perceptible)

• Comparison to traditional graininess metric

– Good agreement between VNHF and graininess metric

• n samples where it can be expected

Comparison to visual demerit ratings

• Visual demerit ratings

– Performed by “expert panel” – Overall evaluation of the 10 color patches – 0 represent “perfection”, larger values are worse

• Analysis based on metric

– VNHF = RMS of deviation from mean of the high-frequency un-structured image – Aggregate over the 10 patches taken as the RMS of the 10 VNHF values – S-curve fitted to map to demerit scale

…comparison to visual

• 4

4 8 11 15 4 8 11 15

y = 1.0177x - 0.7721 R² = 0.9358

Visual Demerit Rating Predicted Demerit

S-Curve(VNHF) Linear.(S-Curve(VNHF)) Other than noise

Pictrography Epson PM750 photo mode Dai-Nippon Xerox DC40 Apple CLW Xerox C55 Ink jet, plain paper, draft

Summary

• Applications for engineering diagnostics

– The metric can automatically separate key factors that contribute to an overall measure of micro uniformity

• Application for competitive benchmarking

– Results are promising with respect to prediction of

• verall visual micro uniformity rating, by combination
• f structured and unstructured variations