The following is a guest post by Barry Wheeler, Digital Projects Coordinator, Office of Strategic Initiatives.
In Part 1 of this series, we examined how a simple scanner was built and showed how the manufacturer determined their claimed resolution. We noted that the International Standards Organization calls this the sampling rate and defines resolution differently. Then in Part 2 we discussed why the manufacturers claimed sampling rate was much higher than the ISO defined resolution and examined images of test targets illustrating the issues involved. We concluded that we should begin with a resolution of 300 dpi or 400 dpi and examine a series of images to determine the best way to operate our scanner (or camera.) But in parts 1 and 2, we limited our discussion to paper documents that were to be presented at the same size or smaller than the original.
Now in Part 3 we are ready to tackle the question “What Resolution Should I Use?” when scanning film or documents that will be enlarged. This becomes an especially important question when scanning our photographs especially from negatives or slides. This post will describe the two major issues that govern our choice of resolution for film or enlarged documents.
Lets tackle the problem of enlargement first. When we scan a document at 400 dpi, we can enlarge it by spreading the dots but it doesnt take much enlargement before the dots and the blank space between them become obvious. All modern image enlargements add new dots in between the originals, guessing at the tones and colors for those new dots. But as the dots themselves are enlarged, they begin to look square because scanner and camera pixels are square! Higher resolution minimizes these affects.
Here is a typical family picture scanned (actually photographed) at two different sampling rates:
Hard to tell the difference isnt it? On screen presentation at a slightly reduced size shows that high resolution isnt always helpful. Now lets enlarge a section from each capture. NOTE: I have enlarged and adjusted the images to enhance the differences. The differences discussed are readily apparent upon close inspection of the images on a high quality monitor.
In figure 3, the square pixels are immediately apparent. Also the enlargement created a halo around the womans hair There is a lack of detail in the eyelashes. The enlargement is about 12x.
In figure 4, the extremely high resolution captured the photographic grain which was enhanced in even this slight enlargement. There is additional detail captured in the branches and hair, but it is somewhat obscured by the grain. The grain on the forehead is obvious and probably unwanted. In this case we might want to scan at a lower resolution so the unwanted grain isnt captured.
Before we discuss enlargements further, lets turn to the second issue. Negatives and slides transmissive materials where the light shines through the images may contain more detail than paper documents where the light is reflected off the surface. In many cases the additional detail adds little or nothing to the meaning of the image. In fact, the additional detail will be invisible on a computer monitor unless the image has been greatly enlarged. Capturing the additional detail makes the image size explode – which makes storage and management more expensive and web delivery painfully slow. Look at the two captures below.
Again, I believe it will be difficult to see much difference despite the large increase in sampling in the second capture. Now lets enlarge a section from each capture. NOTE: Again I have enlarged and adjusted the images to enhance the differences. The differences discussed are readily apparent upon close inspection of the images on a high quality monitor.
In Figure 7, The 400 ppi capture only hints at the detail in the highlight lens. The highlights and texture on the chrome surface have insufficient detail and crispness to give a solid feel of substance.
In Figure 8, the 2400 ppi capture has resolved the detail in the headlight lens as well as enough detail on the chrome and with enough crispness to give a feeling of substance. However, this detail is revealed only under high magnification.
In Figure 9, in this second crop from the same 400 ppi image, the line through the decorative motif on the billboard is not fully resolved. In Figure 10, although the crop is not identical, the line through the decorative motif on the billboard is fully resolved.
Now, after reviewing these images we can return to our question, What resolution should I use when scanning negatives, slides or paper documents that will be enlarged? Again, beginning with enlargement commonly printers perform best with input files of between 240 and 360 ppi. Using 300 ppi as a starting point:
For a 5 x 7 print we would like approximately 5 x 300 = 1500 pixels on the 5 side:
- scanning a 3 ½ x 5 print would require resolution of about 1500/3.5 = 428 ppi.
- scanning a 35mm negative (1 on the short side), about 1500/1 = 1500 ppi.
For an 8 x 10 print we would like approximately 2400 pixels on the 8 side:
- scanning a 3 ½ x 5 print would require resolution fof about a 685 ppi.
- scanning a 35mm negative (1 on the short side), about 2400 ppi.
Capturing all the detail in a negative leads to a new set of answers. The Library has been digitizing some of our most valuable negatives those of the Farm Security Administration photographers taken of the Great Depression in the 1930s. Many are on silver nitrate film which is deteriorating so we know that is the last time they can be captured. Thus we want to capture every information detail (but not the grain) in the negatives. We have used sophisticated image processing software to measure the detail in extremely high resolution captures we find considerable variability depending on the age and type of film, the film base, the quality of the camera and lens, as well as the skill of the photographer. Our estimates may run from 1800 ppi for some negatives taken prior to 1940 to 2800 ppi or more for more recent photographs.
Now consider my change of language! I only know the sampling rate (see Part 1) or the manufacturers stated numbers. We set the camera to sample at 400 ppi or 2400 ppi. But my resolution requirements are based on the ISO standardized resolution of the actual detail visible. I would estimate that our camera was about 90% efficient at a sampling rate of 400 ppi. That would be an actual resolution of 360 ppi. The camera may be somewhat less than 85% efficient at a sampling rate of 2400 ppi giving an ISO standard resolution of about 2000 ppi. Currently a Library contractor digitizes tens of thousands of books from the general collection using a similar overhead camera. But we use a much higher quality and much, much more expensive! camera for our FSA negative.
In the end, our advice is the same as earlier with slightly different numbers begin with a manufacturers sampling rate of around 800 ppi for items that will be magnified 3 to 5 times. For negatives, try to start with a manufacturers sampling rate of 1200 ppi to 1800 ppi. Examine the results carefully if you are happy, start production. If not, be aware that expenses and time increase dramatically with increasing requirements. Then grit your teeth and try a higher sampling rate, or another scanner, or a different workflow, or some final output sharpening. Vary one control at a time and determine what meets your requirements.