9600 dpi. Really?

Evaluation of detail reproduction, based on MTF and SFR curves

In the previous article, I wrote in general about cultural heritage digitization and the importance of image quality control. This paper will be about resolution, sharpness, detail reproduction and the method of its checking.

THE RESOLUTION

The resolution of a digital reproductions is determined by two things:

  • Sampling frequency – the maximum resolution that a camera or scanner can achieve, given in dpi (dots per inch) or ppi (pixels per inch). This is a theoretical upper boundary, and defines the quantity of delivered pixels.
  • Optical and image processing effects – lens design, properties of image sensor, focus, mechanical vibrations, RAW processing settings, sharpening, file compression, etc. This component determines the actual quality of reproduction.

In practice, this means that even if we set the maximum sampling frequency indicated by the manufacturer on the scanner or camera, the reproduction may not contain the required amount of details. We need to evaluate the digitization process, taking into account all its elements, to be sure that we are not creating an “empty resolution”. For institutional-level projects, a digitization with “empty resolution” means a waste of time, money, and other resources.

THE “EMPTY RESOLUTION”

It can be clearly seen that even though we increase the sampling frequency, the system tested in this example is not capable of a resolution higher than 300 dpi.

In this example 300 dpi is the highest value where we can still distinguish all five black lines. Scanning with a higher sampling frequency only produces “empty resolution” which means that the file size increases, but not the amount of reproduced details.

For institutional-level projects the digitization with “empty resolution” is an unacceptable waste of time, money, and other resources.

Source: Image Science Associates: GoldenThread Users Manual

SFR-MTF (Modulation Transfer Function / Spatial Frequency Response)

To evaluate image quality, detail rendering must be measured objectively. With these numerical values it is possible to set up quality requirements and we can also check their fulfillment.

Reproduction can be qualified by evaluating the MTF (Modulation Transfer Function) or SFR (Spatial Frequency Response) curve, and the calculated Sampling Efficiency. This evaluation takes into account all the mechanical, optical, electronic, and software elements of the reproduction process: the camera, the lens, the image sensor, the lighting, and then the settings in the RAW conversion software. Consequently this measurement quantifies how faithfully the reproduction renders the target chart.

The FADGI 4 Stars (preservation-grade) compliant digitization’s sampling efficiency should exceed MTF10 (SFR 10%) > 90%

MTF

The MTF (Modulation Transfer Function) to measure the contrast difference between the original and the digital image, so the MTF is the aspect ratio of the output and the ideal picture’s contrast.

Modulation transfer = Output modulation / Input modulation

SFR

SFR (Spatial Frequency Response) indicates how imaging systems are able to maintain contrast between increasingly finer details.

These functions represents the same properties of imaging systems, but with a different approach – the literature uses this two terms alternately; both be used to quantify the quality of detail reproduction.

Resolution measurement features of the Golden Thread System

In practice, the real resolution, sharpness and detail reproduction of digitization are measured by digitizing the specially developed test chart with the given capture device, and then the finished image is evaluated by a suitable software and the deviation of the reproduction from the ideal chart is quantified.

One of the most common image quality control systems in the international GLAM community (galleries, libraries, archives, museums) is the Golden Thread system of Image Science Associates. This system has multiple features to measure detail reproduction on reflective and transmissive targets.

A Golden Thread Device Level Target felbontásmérő eszközei ©ISa

  1. Slanted square: slanted edges are used for vertical and horizontal SFR measurement and also to check the spatial misregistration of RGB color channels.
  2. Resolution wedges: groups of lines with increasing frequency in horizontal, vertical, and diagonal directions, marked with dpi levels.
  3. Ruled markings (in inches and centimeters): helps to calculate the magnification ratio and the relative resolution in Capture One CH.
  4. Corner gray patches: to check for illumination uniformity.

The method and its scientific background

The purpose of the analysis is to determine the limiting resolution of the reprographic system.

On the resolution test wedges, there are 200, 300, 400…etc. lines per inch, marked with their dpi levels respectively. The cross-sections of the wedges are evaluated by a software by comparing the ideal brightness values of the reference wedge with the values measured on the reproduction.

The cross-section of black and white stripes on the theoretical, perfect wedge has a brightness value of 1 or 0. This generates a square signal.

On a reproduction, due to the imperfections of the reprographic system, the contrast decreases, so that the square signal will be distorted to a waveform.

Source: Image Science Associates: GoldenThread Users Manual

When tested at higher resolutions, the difference between the black and white brightness values will become smaller and smaller (i.e., the contrast will decrease), and finally, at the maximum resolution that the system can reproduce, the wedge bands will completely merge.

From the amplitude of the wave signal at different resolution values, the detail reproduction capability at given resolution can be read and quantified.

Source: Image Science Associates: GoldenThread Users Manual

The measured contrast values are normalized and transformed into a curve by the evaluation software using a mathematical function, the Fourier transform. Thus, we obtain the MTF/SFR curve, which characterizes the detail reproduction capabilities of the reprographic system.

The next figure may make it even easier to understand the MTF curve and its different behavior at low and high frequencies.

Source: Tim Zaman: Development of a topographical imaging device for the near-planar surfaces of paintings, MsC Thesis

FADGI guidelines specifies an expected value for the sampling efficiency at low and high frequency lines. And, of course, there are also different requirements for different types of originals.

A detailed description can be found in the public documentation of the standard.For example, FADGI 4 Stars sets the target resolution for reflective photographic materials (photos, enlargements) at 600 ppi. It accepts a range of 40-65% of Sampling Efficiency for medium frequency (MTF50) and >90% for high frequency (MTF10) details.

The requirements for transmissive photographic materials (negatives, slides) are similar, except for resolution, which is 3500 ppi[1] in this case, but other acceptable ranges are the same: MTF50 40%-65%, and MTF10> 90%.

Interpretation of the MTF curve with Golden Thread software

Source: Image Science Associates: GoldenThread Use and Interpretation Manual

There are three notable points to observe on the curve:

  1. Amplitude values >1 – indicates the degree of over sharpening
  2. Nyquist frequency – the amplitude read at half the target value
  3. Limiting resolution – the value where the contrast drops below 0.1

The tolerances and limits required to achieve the various FADGI ratings are clearly visible.

For comparison, the MFT analysis of an inappropriate negative-digitization. It is visible that despite the 4000 dpi target resolution setting, the actual rendered details are 1000 dpi (horizontal) and 2000 dpi (vertical).

Source: Tepper, Aap: Mass Digitisation Quality Control Workflow: Photographic Negatives (Film Archives, The National Archives of Estonia)

Conclusion

Measurement and analysis of MTF (Modulation Transfer Function) and SFR (Spatial Frequency Response) is a scientifically recognized method for evaluation of detail rendering and image quality control in reprography because:

  • objective, thus precluding subjectivity arising from visual assessment,
  • takes into account all elements of the digitization process,
  • the measured values are comparable, to determine compliance with different standards,
  • the quality of digitization of both reflective and transmissive originals can be examined.

In addition to detail reproduction, the SFR curve quantifies several imaging features, such as sharpening, aliasing, color misregistration, and overall image quality.

Using the Golden Thread System it is possible to measure the true detail rendering capabilities and the image quality of digital reprographic systems and verify the compliance with international image quality standards.

Resources, links, further readings

Contact

If you have any further questions, please reach out to me through the contact form:  CONTACT

Daniel Horvath
freelance photography technology consultant

[1] The tolerance of 4000 ppi for transparent originals in the 2016 text was extended to 3500 ppi in 2019 at the suggestion of Don Williams.