1) Preserve the images in as high a quality as practical using current technology.
2) Use a format that was as "future proof" as possible, to try and ensure that the files can still be read for as long as possible. There are many examples of the latest technology being used for projects, only for the content to be unreadable in a few years because the technology became obsolete or unsupported (see BBC Doomsday project as an good (or bad!) example.
http://www.guardian.co.uk/uk/2002/mar/03/research.elearning
http://en.wikipedia.org/wiki/BBC_Domesday_Project )
3) Store as much metadata (information about the content of the images) as practical in a way that it will always be associated with the image, irrespective of where it may be copied.
After much research we decided upon uncompressed lossless TIFF (Taggged Image File Format) in 24bit colour or greyscale where applicable of at least 300 DPI (dots per inch) and more where appropriate (i.e. small images). This is a relatively old but mature standard, and although doesn't produce particularly small images (some of our images are over 80 MB), they retain the full original quality, are easy to convert to other formats (i.e. we use JPEG for the web site with 90% compression as a good compromise between size and quality for online viewing and delivery). Copies of images in TIFF format will always be stored as master copies.
We also decided to use the IPTC (International Press Telecommunications Council, www.iptc.org) Photo Metadata Standard to store information within the image files themselves. Again this is a well respected international and mature standard. This enables us to store a multitude of attributes for each image such as source (who supplied it, who holds the copyright if any), where the original can be located (Hilton Digital Archive or elsewhere), and probably most important, information about the content of the image (who, when, where, what) if known. The advantage of embedding this within the image file itself, is that it will always be retained with the image irrespective of where it may be copied, and can be read with appropriate software.