`3. Conversion

Key Concepts

scanning factors
rich digital master

  continuous-tone   halftone
proposed method guidelines

additional reading



Contents Selection- intro Intro- digital images Conversion-benchmarking stroke Quality Control Metadata Technical Presentation Digital Preservation Management Continuing Education
Conversion-benchmarking text Conversion-benchmarking continuous-tone

The QI method was designed for printed text where character height represents the measure of detail. Manuscripts and other non-textual material representing distinct edge-based graphics, such as maps, sketches, and engravings, offer no equivalent fixed metric. For many such documents, a better representation of detail would be the width of the finest line, stroke, or marking that must be captured in the digital surrogate. To fully represent such a detail, at least 2 pixels should cover it. For example, an original with a stroke measuring 1/100 inch must be scanned at 200 dpi or greater to fully resolve its finest feature. For bitonal scanning, this requirement would be higher (say 3 pixels/feature) due to the potential for sampling errors and the thresholding to black and white pixels. A feature can often be detected at lower resolutions, on the order of 1 pixel/feature, but quality judgements come into play.


Stroke: Adequately rendered cloud outline (left) and inadequately rendered border line (right).

Cornell has developed the following correlation of perceived image quality to pixel coverage:

Table: Quality Index for Stroke Rendering

QI Quality Assessment
2 excellent
1.5 good
1 questionable, confirm quality onscreen
<1 poor to unacceptable

Grayscale/Color QI Formula for Stroke

dpi = QI/.039w

This formula correlates QI with dpi and stroke width (w) measured in mm. QI in this case is based on the quality assessment above, which correlates to the number of pixels covering the stroke (e.g., 2 = excellent). Note: if the measurement of w is expressed in inches, omit the .039.

For bitonal scanning, the formula is adjusted to compensate for feature drop out in the thresholding process:

Bitonal QI Formula for Stroke


Many items falling into this category exhibit features beyond simple edge-based representation, and resolution will not be the sole determinant of image quality. For example, a number of institutions have recommended scanning all manuscripts in grayscale or color.

Reality Check

What's the minimum resolution I need to scan a manuscript page in grayscale to obtain good rendering of the finest stroke, which measures .1 mm?



© 2000-2003 Cornell University Library/Research Department

Conversion - benchmarking textConversion - benchmarking continuous-tone

View this page in Spanish
View this page in French