Silver Halide Resolution

I saw a page at Hewlett-Packard Photosmart web site (in mid 1998) saying that typical 4" x 5" silver halide photo would only carry about 200 DPI resolution (!) and due to that there would be no reason to scan photos at higher DPI than that. I can not find that page anymore at H-P site, but this "information" has been actively adopted by many people, there are many web pages that distribute it and every now and then someone claims the same on the Usenet.

The 200 DPI sound so suspicious so I had to inspect it a little.

The original photo was glossy Agfa paper, it was processed by a typical print-shop some 4 years ago. It was taken with Olympus AZ1 point-and-shoot pocket camera onto Fuji DS 100 ASA color negative.  So there is nothing high-tech in the process flow or with the camera. It is typical snap-shot. The red arrow shows the location of inspection #1: How much detail is available in typical silver halide snap-shot photo. The yellow arrow shows the location of inspection #2: What is the resolution potential of common snap-shot silver halide photo. (The model is one of our birds, Tipi)

How much detail is available

.

Please click the thumb on the right to see the optical enlargement in full-size (taken with Canon EOS*DCS*3 + FD Bellow + 35 mm lens). The on screen magnification is about X120 depending on your screen resolution.  The 1 mm scale (at the bottom) is photo of a calibration scale taken with the same setting as the target image, it has 10 micrometer division so it show that this imaging configuration can easily resolve 100 lines/mm that is comparable to 5080 DPI but actually it can do at least two times of that, I just do not have any denser grid to measure it out.

The enlargement appears to be noisy but it is not noise. The "noisy appearance" depends on the lighting situation and properties. Polarization will remove it but it removes some data also so I choose not to filter it out using polarization.
 
Now how much image data there is? At 200 DPI one dot is 127µm per side (as much as 12.7 divisions from the scale) and there clearly is much more data to capture. Something like 1000 DPI (2.5 divisions) seems to be sufficient for capturing most, but not all, of the data. Remember that this is a low quality snap-shot photo, better camera/lens alone can certainly provide even more data. 

What is the resolution potential

This was not an easy one to inspect as can be seen from the image below, it is not exactly such a high quality image I would like it to be but it does show something interesting so I have put it up and will try to improve it later.

The silver halide photo does not like high intensity light from the microscope at all, the dyes or the fixation actually moves due to the heating so the region of inspection jumps out of focus after a second ot two. I had to stack all my IR filters into the light path of OM-BH2 microscope. 

When using the common coaxial bright-field lighting only the scratches at the topmost layer of the photo are visible, the layers beneath show only as an even murky brown surface. I had to use polarization. 

Next thing to notice was that the dyes are in different layers and with a 100X dry objective I had not enough depth-of-field to capture all the cyan, magenta and yellow layers simultaneously, only one layer was sharp at the time. I had to take three images and compose them in Photoshop.

Next problem was that the photo did not like the tungsten color temperature, colors were somewhat indistinguishable and blurry. I had to add a red-cut filter (Tungsten "daylight" filter) into the light path. 
 
And finally due to the various filtering I had so little light that it was very difficult to get focussed and the light was not nearly enough for the camera, so I had to use long exposure combined with noise reduction. I ended up with 30 images, 10 for each layer for noise reduction. The resulting three images were then combined by cyan, magenta and yellow. 

The on-screen magnification is around 2700 (imaging configuration was OM-BH2 microscope + X100 dry lens + 3.3 photo-ocular + Canon EOS*DCS*3 camera + heavy filtration). The scale is (partial) photo taken with the same imaging setup as the target image from the 1 mm calibration scale that was used also in the inspection #1 above.

So, what does this tell. The dye particles are extremely small, in the range of 1µm and below, so a 20µm dot can hold some 400 such dots per each primary color. This is far more than the 8-bit gradation (24-bit color). 

20µm is the same as 1270 DPI, so it seems that scanning at 1200 DPI using higher bit-dept like 12-bit (36-bit /color) will capture most of the details from silver halide photo.

Both the experiments shows that the full detail resolution of a silver halide print is quite high, something like 1200 DPI or more at higher bit-depth than 8-bit even for a photo from a snap-shot camera. 
 
According to this one should be able to digitally enlarge a 4" x 5" photo by scanning it at high DPI such as 1200 DPI or 2200 DPI without loss of resolution. Even for a high resolution 300 DPI dye-sub the no-loss magnification would be as much as X4 or so. This is inline with my experience with such magnifications.

How Does this Relate to the Resolution of 35 mm Film?

For the 7" edge of a silver halide there is at least 1200 dots of real information for each inch, 7*1200 = 8400 dots. 
 
In order to have that accuracy from the 35 mm film, the film scanner must be able to get those 8400 dots from the film. The long edge of the film is 36 mm = 1.42 inch so a film scanner of 8400/1.42 = 5915 DPI is required.
 
The above is only the spatial resolution. The data (image) in film is highly compressed so in order to get it out from the film, a very low-noise film scanner that has high bit-depth (14-bit or more) is needed. On the silver halide the data (image) is in near linear space so 12-bit is sufficient for capturing it.