With professional cameras sporting over 39_MPixels, Photograpy said "Good Bye" to the analogue age. Today, most cameras sold are digital and in a few years everybody will own several ones. One big advantage of storing pictures in computers is that there is no cost associated with taking a shot, other than recharging the battery. The future is an ocean of snapshots.

These developments, however, beg for the question of how are people going to find a certain image from the zillions stored around them? This implies actually two issues.

The more difficult one is related to how humans store images in their memory and how is it possible to formulate (describe) that internal representation in terms of the "physical" image? Let us call this the "query protocol" problem.

Google and other search engines use text tags found close to or directly related to the image for performing image search via text-search. Hence, the query is formulated here in textual format. By contrast, content based image retrieval (CBIR) finds images based on their "similarity". The query is now an image and the system must find and rank all images similar to the query from a very large database. This is the "content addressable protocol" problem.

In the near future - as hardware quality continues its exponential growth - the CBIR protocol will gain terrain. However, there will be no real solution to the image-search task, unless a very accurate and robust image and image-object-scene representation in terms of simple features is found. The AMASS platform will excel at searching in a content dependent, robust way such feature-arrays, coding for the content of images AND of the correlated text. For example, the AMASS platform could perform easily the retrieval part for the biometrical iris-recognition, because taht problem has a simple and effective representation in terms of binary features.

Searching digital videos might be actually simpler, because additional cues - like text, audio-track, camera motion events - are also available.