Measurement of digital camera modulation transfer functions (MTFs) using sinusoidal test patterns shows that signal processing provides a high MTF value until the spatial period of the test pattern matches the pixel size, at which point the MTF falls rapidly to zero or the test pattern lines are distorted by aliasing. Let's designate this value the spatial frequency bandwidth (SFB) of the camera measured in the image plane. This number is equivalent to the resolution limit in film photography.
Sector star test pattern (a) and its image (b) at x1000, D being the unresolved area of the test pattern.
We need to choose an optimum print width (OPW) with the detail of our camera (SFBc) that matches the detail that can just be seen by eye (SFBe) when, for example, we view the print at arm's length. We can think of this as matching the size of the cones in our eye to the size of the pixels in the camera. We must be sure that our printer does not degrade this resolution. Since there will be the same number of limiting cycles in the image as in our print (assuming no cropping), it follows that OPW = d(SFBc/SFBe) where d is the maximum width of the camera format. The value of SFBe is approximately 5 c/mm. We can measure the SFBc values using a sector star pattern with 36 sectors (see figure) filling the screen of a laptop so as to generate an incoherently illuminated test object. If D is the diameter of the unresolved area in the image of the test pattern, then SFBc = 36/πD, in which case OPW = 2.3d/D. Although the ratio d/D can be found from the test conjugates, it has been suggested to me by Tom Williams from Sira Limited (Chislehurst, UK) that it can be obtained directly from the scale readings in Photoshop or equivalent software. If we record this pattern at a range of 1000 times the focal length of the camera lens, when we load the image back into the computer and set it to a size equal to the test pattern, we will see an unresolved gray disk of diameter D (mm) at the center. This value together with the full width of the image format, d, can be read off from the scale in Photoshop, enabling the calculation of the OPW. This test could also be carried out using a print of the test pattern rather than a computer display if more convenient.
Results obtained on 10 models of digital cameras having pixels in the range of 1 to 4 megapixels gave OPW values in the range of 90 to 180 mm. These values would be halved if the print were viewed at the least distance of distinct vision (250 mm). The same test carried out on a zoom film camera, using a microscope to measure D, gave an OPW of 400 mm at wide angle down to a disappointing 90 mm when used at the longest focal length. Digital cameras with a much shorter focal length seem to perform better in this regard.
The OPW provides a useful parameter for camera spatial image quality because it takes into account effective pixel size, the number of pixels, lens quality, and the effects of signal compression. It does not, however, cover distortion or color reproduction, although further experiments using software such as Photoshop over different color bands may prove of value.
We have set up an informal digital camera testing group to exchange results for different models with the objective of agreeing on a simple, meaningful test. For more information, send an e-mail to the address below. oe
Lionel Baker is a consultant in Orpington, UK.