• Federal Agency Digitization Guidelines Initiative (FADGI) – a US based interagency government effort
• METAMORFOZE – a venture between the National Library and National Archive of the Netherlands.

These are known, in shorthand, as FADGI and METAMORFOZE. In general, their similarities are much greater than their differences. Many institutions have adopted these guidelines for their own in house digitization and carefully scrutinize hardware, software, and workflows to ensure they meet or exceed the requirements set forth in these guidelines.

Both FADGI and METAMORFOZE describe several tiers of quality. Loose equivalencies of these tiers are shown below.

*In the official documentation the three tiers are unhelpfully referred to simply as “METAMORFOZE”, “METAMORFOZE LIGHT” AND “METAMORFOZE EXTRA-LIGHT”. Here we are taking the liberty to use “Strict” to refer to denote the “METAMORFOZE” tier.

The purpose of these tiers is not to pass negative judgement on digitization executed at less-than-preservation-grade quality. There are some use-cases where there is no incremental value in accomplishing digitization at a higher quality. For instance, neither color nor tonal-accuracy is critical for patron requests for text-only circulation material; if the patron can read the result clearly, the quality was more than sufficient for the task.

However, as we argued in Unforeseen Costs of Imaging for Short Term Needs we feel strongly that Preservation Grade Quality (METAMORFOZE-Strict or FADGI 4-Star) should be pursued unless it’s eminently clear that such quality will not offer additional value.

“The principals behind FADGI and Metamorfoze are exactly the same: measurable and standardized guidelines for assessing digital imaging performance. They both use the exact same ISO methods for measuring performance and both use graduated (or tiered) levels of imaging performance specification. The differences lie in the flexibility and choices of those specifications. Metamorfoze was intended for general preservation imaging tasks and the specifications are designed for such a use case with three distinct non-interchangeable categories. FADGI guidelines can also be chosen for preservation but also allow for lesser or greater needs. For instance, color imaging performance in FADGI can be specifically relaxed relative to other performance metrics ( i.e. interchangeable) if the use case supports such a choice. There can be a greater variety of performance levels between imaging metrics in FADGI than in Metamorfoze. Generally though, FADGI three star levels are the same as all Metamorfoze specification levels.”

– Don Williams, Image Science Associates

Technical Specifications of FADGI & METAMORFOZE

Resolution: Beyond “Optical” vs “Interpolated”

Sampling Frequency is the number of pixels (typically measured in ppi, or pixels per inch) used to represent the physical object. This measurement does not represent how much meaningful detail is present; it is unaffected by inherent sharpness or additional sharpening added in post processing.

Imagine that a professor asks his students to write a 20-page essay, and every student in the class turns in an essay that fills 20 pieces of paper. In the strict sense, all students have fulfilled the requirements of the assignment. However, some students use double spacing, others add padding between the letters, and many use filler words and phrases that add no actual detail to their arguments, and only a few actually fill 20-pages with precise and meaningful content.

This example is very similar to the way that many, perhaps even most, digitization systems claim to scan objects at a particular PPI; at closer inspection the amount of precise and meaningful image content is well below that PPI. Worse, over time such marketing misrepresentation of the actual detail level has become better hidden.

For many years, Cultural Heritage Institutions have known to look for “optical resolution” in scanning systems, as scanner manufacturers would often provide non-optical interpolated resolutions in their marketing material and only show their optical resolution deep in technical specification lists. This is the equivalent of double-spacing an essay, and is quite obvious and well-known. Here are more subtle forms of marketing inflation:

• Theory vs. Measured: Stating only the resolution of the sensor, which provides a theoretical maximum resolution, rather than the as-measured resolution of the entire system. A 36 megapixel sensor used with a mediocre lens does not record files with anywhere near 36 megapixels of actual subject detail.
• Center vs. Edge: Stating the resolution measured in the center of the image, rather than the resolution in the corners or edges. Most general purpose photographic lenses, including even expensive, high-end lenses, are not as sharp in the corners as in the center.
• One Axis vs. Both Axis: Stating the resolution measured in one axis, rather than both. Scanning systems (planetary, flatbed, or rotary fed) that move during the capture of the image often have measurably better detail along the axis of the scanning array than along the axis of their motion. In other words, the image is subtly blurry in the direction of the motion.
• Sharpened vs. Native: Many systems have excessive sharpening applied automatically or in the background, hiding the native sharpness (or lack thereof) from the user. The resulting image appears visually sharp in the same way an auto-tuned singer sounds on-pitch.

As manufacturers exaggerate resolution in increasingly sophisticated ways, it has become clear that the community needs a foolproof, objective way of measuring resolution. The members of the FADGI Still Image Working Group assessed many proposed measurements for the level of actual subject detail recorded in an image, and settled on Spatial Frequency Response (SFR).

SFR is a fairly complicated mathematical modeling of how well detail is recorded at various feature sizes. It has many features, including the ability to reveal when an image has been over-sharpened, and allows a more holistic analysis of detail than any metric based on a single number. Simplicity, however, is not among its advantages. The formula used to generate a graph, which is then analyzed to provide the final relevant rating number, is available on the Wikipedia article on Optical Resolution. It is unlikely that any institution would undertake manual calculation of SFR. Instead, SFR is almost always calculated by one of the several image-analysis software packages discussed in Numerically Evaluating Image Quality.

The bottom line is summarized well at Archives.gov:

“Do not rely on manufacturers’ claims regarding the resolution of scanners/digital cameras, even optical resolution specifications are not a guarantee the appropriate level of image detail will be captured. Most claims are over-rated in regards to resolution, and resolution is not the best measure of spatial frequency response (modulation transfer function is the best measurement).”

– (page 33, www.archives.gov/preservation/technical/guidelines.pdf)

Tone & Color Accuracy

The ability to compare objects from disparate institutions, digitized at different times, and potentially digitized by different equipment, requires very precise control of tone and color. Both FADGI and METAMORFOZE specify tone and color control by providing tolerances for:

• Maximum Delta E: All measured color-errors must be below this value.
• Average Delta E: The average color-error cannot exceed this.

FADGI and METAMORFOZE use different formulas:

• Delta E 1976 | METAMORFOZE: The original Delta E formulation.
• Delta E 2000 | FADGI: A modern revision that weights errors based on how likely it is to see a difference between two colors assuming the intended user of the images is human (as opposed to machine analysis).

The limits for tone and color error are given below:

Noise

Both FADGI and METAMORFOZE suggest the measurement of noise, but do so in slightly different ways.

• FADGI: Standard deviation for each neutral patch plus red, blue, and green patches.
• METAMORFOZE: Standard deviation in each patch.

The limits are given below:

The measurement of noise is especially prone to three complications:

• Noise Reduction in Software: Many digitization systems apply noise reduction by default, or in a way that is not transparent to the user.
• Unclean Targets: Foreign matter such as hair, dust, or grime will often be read as noise. A very clean target is required for proper analysis of noise.
• Real-World Variation in the Target: The assumption in the formulas is that the subject matter is, in reality, 100% homogenous, so any variation at all is a capture error. In practice, however, even the highest-end targets can have minor imperfections.

Illumination Uniformity

Both FADGI and METAMORFOZE provide methodology for evaluating the evenness in which the object is illuminated. This can be accomplished natively or by post processing compensation. The latter technique opens up a host of possibilities, including:

• Allowing for a smaller working area than could otherwise be achieved.
• Allowing for non-even source lighting that enhances texture without compromising uniformity of the subject’s brightness.

Major limitations and challenges to measuring or achieving illumination uniformity include:

• Bound Materials: When a bound material is imaged at a non-180 degree opening, one page will inevitably bounce light into the facing page; when there is any variation in the density or color of the page (e.g. a heavily illustrated page, a manuscript with gold embellishments, or a blank page) then this will result in a variation of the illumination uniformity. This can make it nearly impossible to achieve FADGI 4-Star illumination uniformity in bound materials, unless a different correction is made and applied to every individual page.
• Only Luminance is Considered: Some systems may have slight color biases in one area of the frame. For instance, when illumination is provided by two lights on opposite sides of the object, a slight difference in color temperature between the two light sources will manifest itself most at the outer edges of the frame. If the target patches used to measure color accuracy reside in a small part of the overall frame, then this drift in color will not affect color accuracy, and may not be detected by Luminance Uniformity checks based only on luminance.

Color Registration

Low quality lenses and systems that use motion for capture (e.g.: linear and planetary scanners) can cause misregistration of color. Such misregistration can manifest itself as a ghosting-offset or chromatic aberration, and is more commonly seen as a color alongside hard-edge, high contrast lines. Misregistration of color makes it difficult to judge small, fine details, as it is impossible to know (especially in the context of a future viewer separated from the digitization process) whether such offset was present in the original (e.g. a production run of newspaper where the printing press plates were misaligned) and correctly captured, or whether it was an artifact of the digitization program.

Next Page: Visually Evaluating Image Quality →