ISO 12233 is an international standard within Photography: Electronic Still-Picture Cameras, titled "Resolution measurements". The first edition was published in September 1st, 2000 - hence named ISO 12233:2000.
It defines terminology, test charts and methods for performing resolution measurements on digital still-picture cameras. In practice this standard is also used for various types of cameras, including linear scanner cameras.
Generally the specifications can be split up in four parts. The first describes a target that is rarely used in professional photography containing dozens of subtargets designed to assess resolution using multiple methods. Furthermore, test conditions, measurement methodology and presentation are discussed. A summary with key standards are discussed in the following sections. For details and more specifications, acquire the official document.
The test chart should be surrounded by (matte) black surroundings with uniform illumination. Baffles between the light sources and the camera to prevent direct illumination. Clipping is not allowed in both ends of the signal, but the brightest white in the target should provide a near maximum signal level. Compression of the image is allowed but have to be reported in the visual presentation.
A test chart should contain multiple slanted edges, in horizontal and vertical direction. Although not stated in the standard, an edge of near 0 degrees (or 90 degrees) will result in too little differentiated data. The same counts for near 45 degree edges. Many slanted edge target types therefore use a slanted edge of 5 degrees.
- The indicated algorithm is divided into 6 parts.
- Convert the digital signal values to a linear quantisation characteristic (using OECF ).
- The image should be derived/convolved with a [-0.5 0 0.5] filter window. For each line, the edge will be represented by a maximum.
- A line should be fitted over the centroids of the previous maxima.
- Compute the line spread function (LSF) by shifting all the lines (or rows) in the image by an amount equal to the shift of the line from the first row. If we want to use a Discrete Fourier Transform (DFT) later on, we have to discretize our LSF with equal steps. Therefore, we can bin our data into four bins per pixel width.
- A Hamming window can be applied to the LSF in order to reduce noise effects or any other information away from the edge.
- Lastly, a Fourier transform is applied to the windowed LSF. The SFR then follows if we normalise the modulus of the previous Fourier Transform.
Influential camera settings have to be displayed, including lens focal length, aperture, resolution and compression mode (if applicable). If dark/flat-field correction was used. Illuminating source colour temperature and level. Spatial values should be reported in LW/PH. The MTF graph should consist of two lines, representing the horizontal and vertical SFR. These graphs should be derived from 8 SFR measurements, 4 from 'black to white' and 4 from 'white to back'.
This guideline can be used to derive the SFR of several targets, among the following:
- Applied Image QA-62 (supported by Deltae)
- Golden Thread Target (supported by Deltae)
- Universal Test Target (supported by Deltae)
This is a common standard in resolution measurement. Many (software) applications state the use and implementation of this standard in their algorithms, although it is usually not strictly followed. For example, many applications use the separate channels [R,G,B] in their visualisation, instead of one weighted line per orientation.
It demands a minimum modulation between the dark and white edge of 20%. In practice, (digitally) changing just the relative strengths of these two can influence your MTF values to some extent.