A report describes progress in a research program oriented toward developing artificial-intelligence capabilities for processing images and sounds in changing environments. This research focuses on extending the state of the art in mid-level processing of visual and auditory signals, following a biologically inspired approach. Developments described in the report are summarized as follows:

  • In psychophysical experiments, Kalman filters were shown to be useful for modeling hitherto-unexplained features of adaptation of the vision systems of human observers to changes in speed.
  • There was a continuation of previous research on low-level features (and processing of the features) needed for detection of junctions (e.g., characterized by L-, T-, and Y-shaped intersections on objects), which are known to be critical for classification of objects.
  • Another continuation of previous research included refinement of lowlevel features (and of processing the features) needed for classification of sounds in the presence and absence of noise. These features and the processing thereof are closely related to those of visual processing.
  • There was a continuation of development of an integrated perceptual system that would combine low-level feature extraction, attentional mechanisms, and simple object recognition to control a robot arm engaged in a task.

This work was done by Bartlett W. Mel, Norberto M. Grzywacz, Laurent Itti, and Shri Narayanan of the University of Southern California for the Naval Research Laboratory.


This Brief includes a Technical Support Package (TSP).
Biologically Inspired Processing of Image and Sound Data

(reference NRL-0022) is currently available for download from the TSP library.

Don't have an account? Sign up here.