Handling digital photos is very different from analogues. CeWe Color found out that after having taken, saved on the computer and quickly examined their photos, many people find themselves overcharged with decisions like “which ones to take for prints” or “which one use for a poster for the grandparents”. Taking into account that the amount of digital photos triples the amount of analogue photos taken, one can easily imagine how long it takes to sort out the photos of a two weeks holiday. In the past all photos where printed before it was decided to “some day” put them into an album. With digital photography the selection process is usually done before printing so that about 80 % of all photos never get printed and reside in “digital shoe boxes”. A great business potential lies in convincing the user to open this shoe boxes by providing him with advanced digital photo services. A prominent example is the CeWe digital photo book which can be authored on a computer and be printed at CeWe Color in premium quality. Photos can only be a way to capture ones memories when they are looked at. What is needed are services that immediately bring people to the point of using them without great efforts.

One way to achieve this is assist the user in selecting photos. Imagine having beeb on a holiday trip and having taken about 700 digital photos. When

preparing a nice photo book as a means to preserve the memory to this event one rarely wants to have all of these photos in the photo book but selects the photos that best represent the overall holiday trip. Manually selecting selecting these photos often is a very time-consuming and tediuos task. Automating this task for their customers would be a way for CeWe Color to convince them to use their services and not the ones from competitors. In PickEasy methods are developed fulfil this goal. For this image analysis as well as context analysis techniques are employed to decide if a photo is good or bad. Doing this is not an easy task and often cannot be decided on analysing single photos. One can for example decide if a photo is unsharp by employing edge detection on the photo or if it is under- or overexposed. But if this photo was the only one to represent a certain event one would probaly select it despite of its poor quality.

Taking this background into account a prototype system is developed that semi-automatically selects the "best" photos out of a large set of photos. Therefore a component-based architecture is being developed that combines several content- and context-analysis techniques and iteratively enhances photos with metadata. On the basis of this metadata intelligent methods are being developed to retrieve photos from an input photo set. This selection can be used for prints but also can act as input for the semi-automatic generation of digital photo books. Currently the results of this research activities are directly employed in the CeWe photo book application.