Adam's Image Transfer Page

A. N. Other visitor writes:

> keep up the good work, by the way how did you get such great shots on
> your page?  What kind of hardware is involved?

I'm glad you asked. Let me describe ...

Step Zero: Acquisition

Take good photos in the first place! We use the ubiquitous Canon EOS, we've an EOS 1000FN and an EOS 50. (and a Canon digital camera) Our lenses (a 24-80mm zoom and a 75-300mm zoom) aren't that great, and we use ASA200 colour print film, but we still get OK results. We used to get 1-2 good photos per roll and now we're upto 3-5 so we're either getting better or our standards are dropping!

Step One: Transfer

Many of the pictures are from Kodak PhotoCD, a process which scans direct from the slide or negative at high resolution (although professionals would still call it medium resolution) and very good quality. We get our films developed and printed as normal, then choose the best ones and take the negatives in to be PhotoCDised. Within a couple of weeks they come back on a beautiful gold CD. You can get 100 pictures on a disc and it's "multisession" so you can do it in several goes - you don't have to do 100 at once. It's a bit scary sending your 80% complete disc off though :-)

The images are on the CD in a proprietary format, each in an "image pac" comprising five resolutions: Base (768x512), Base/4 (368x256), Base/16 (184x128), 4Base (1536x1024) and 16Base (3072x2048). Most of the images on my homepage are from Base reolution images, but a couple are small areas of 4xBase. The quality is excellent - much better than on print film. I have a couple of well-focussed images where you can see amazing detail in headlamp/sunglass/eye reflections.

Sometimes I scan prints using a normal flat-bed scanner, which is OK at web reolution, but does lose contrast. Then there's negative/slide scanner, which is better, and of course digital camera.

Step Two: Processing

The scanning/transfer results vary. When you take your exposed film to a D&P lab, there are quite a few variables which determine how your prints will appear. The first, of course, is the exposure the camera gave to the scene. The second is the exposure the d&p stuff gives to the print. Finally, there are colour balance settings on the machines to take account of different film's responses. So the lab has a big influence on how your prints look. You can usually under- or over-expose your shots by a couple of stops and the lab will "correct" it for you. Of course, whether they're correcting or ruining the shot is a matter of interpretation.

Our Kodak developers get good results from Kodak film but Fuji film comes out very green (I suspect this is because they just leave the machine on the "Kodak" settings). I think the Photo CD transfer is done as a very "dumb" lab - they don't seem to do any colour adjustment or exposure "correction" to the transferred image. So when I look at the scans I often have to do a fair bit of mangling to get the brightness and colour right. The Grand Canyon pictures (taken on Fuji film) needed a lot of reddening; almost all of the sunsets needed a lot of darkening, etc. The Golden Gate Bridge pictures were taken at night but the camera exposure was almost like daytime! The print most closely resembled our memory, but the Photo CD transfer was faithful to the negative.

Having too much light isn't necessarily a bad thing, however. If a print is too dark, you can lose detail in the shadows, where many shades collapse to be black: When you lighten the image, those collapsed shades cannot be separated, and just go to a single, lighter shade of grey. If there's more light than you need, you get the detail from the shadows which you can darken if you choose. Of course, you need to make sure that detail isn't lost with highlights collapsing to white ...

I agonised for a while about whether it was "cheating" to manipulate the images in that way but in the end I decided it wasn't. Sometimes, like in these sunsets, it's hard to know what the "real" image should be. It's definitely not the raw scan, but is it just a case of matching the print, or should I tweak it until it most closely matches my memory? I decided that it's OK to do some manipulation, as it's just what the lab does to get the prints anyway.

For the manipulation I use the superb Gimp under linux. There are a lot of PD and shareware image manglers which can do gamma correction, colour balance, contrast enhancement etc. so if you can get one to load PhotoCD you're away.

Step Three: Compression

Finally I convert the image to JPEG, using the lowest resolution which doesn't result in a perceptible loss of quality at normal size. Normally this is between 70% and 90%. for a 768x512 picture it can give a JPEG between 50 and 150Kb depending on the complexity of the image. I usually try to keep it under 100Kb if I can, just to save a bit of download time for the viewer.

Progressive JPEG is potentially slightly smaller and more interesting to load from the browser point of view.

Optimising the JPEG compression (using jpegtran) is well worthwhile - images from digital cameras typically shrink by 10% with identical quality.

Step Four: Installation

Then it's just a matter of getting it to my www space, with ISPs you'd typically use ftp, but now I've got my own web server I have a local copy of the webspace and use rsync to keep them in step.

Step Five: Viewing!

Of course, the more pixels and the more colours the viewer has on screen, the better. I would recommend 90 dots-per-inch at 16 bits-per-pixel as a minimum. On a 14-inch screen 1024x768 is nice. On a 21-inch monitor 1152x900 or 1280x1024 would be necessary.

8bpp is pretty cruddy and pictures suffer greatly from the "dithering" which must be performed to fit the quart into the pint pot.

(although to be strictly accurate, fitting 24-bit colour to an 8-bit palette is like fitting eight thousand gallons into a pint pot! No wonder it doesn't look that great)

Going from 8bpp to 16 is an enormous improvement, losing most dithering effects and giving, from a couple of feet away at 90+dpi, near-photographic quality.

Personally, for normal photographs I don't notice the improvement between 16 and 24bpp nearly as much as the jump from 8 to 16. I suppose it depends on the nature of the images, as the jump is easily noticeable for computer art.

These days, 24bpp isn't uncommon. When I started this page in '95, I had an Acorn Risc PC at home with 1024x768x16bpp or 800x600x24bpp (I preferred the 16bpp setting), while at work I used a Sun Ultra 1 with a 1280x1024x24bpp display. By '01 I had a 1600x1200x24bpp display on my laptop ...

Apart from resolution and colours, the other factor is monitor gamma. Essentially this is the brightness of the monitor. Some are bright and some aren't, is the best way of putting it, and my pictures which look great on a bright monitor might look a bit dingy on a dull one. Still, you can always adjust the monitor brighter, but if I darkened the images I'd lose detail in the dark bits ...