Automated Image Search Engines

We mentioned that MPEG-4 codifies meta descriptions so that complex content can be
archived. This work is carried forward into MPEG-7 to provide support for automated
image search engines (equivalent to word search engines). Remember that we are performing
a discrete cosine transform on each macroblock within an image. We are therefore
expressing the spectral content and frequency content (that is, color) of each
macroblock in terms of a series of digital filter coefficients.
MPEG-7 exploits this process to produce a standard for automated content
description—and hence automated content searching. All images are converted into a
common unified format in which image features are identified based on the wavelength
of colors making up the scene. The frequency content is described as a 63-bit descriptor.
Now consider this: A medium-sized town has, say, 20 surveillance cameras that are
taking pictures every second or so. Those pictures are being stored in a database. The
police want to look for someone in a red woolly hat who walked across the right-hand
topmost macroblock of camera number 20 at some time during the past 6 months. Previously
this would involve several police officers looking through endless video
archive footage. The meta descriptor automates this process—just search for red in
macroblock x and wait for the results.
MPEG-7 makes wireless-enabled video surveillance far more powerful, because it
simplifies the image archive search and retrieval process. The bandwidth uploading
from surveillance cameras is generally non-time-sensitive and can occupy the long
low-load night hours. Color depth is also important in these applications—the fact that
the suspect was wearing a red woolly hat. Contrast ratio is also important.