Biologically Inspired Processing of Image and Sound Data

Overview

The document is a progress report for a project titled "Next-Generation Image and Sound Processing Strategies: Exploiting the Biological Model," funded by the Office of Naval Research (ONR). The report covers the period from February 1 to April 30, 2007, and is authored by a team from the University of Southern California, including Principal Investigator Bartlett W. Mel and co-investigators Norberto M. Grzywacz, Laurent Itti, and Shri Narayanan.

The overarching goal of the project is to advance mid-level visual and auditory signal processing by utilizing biologically inspired methods. The report outlines two primary objectives: (1) to imitate biological sensory feature extraction methods, and (2) to apply these biologically inspired features to enhance attention on critical information within complex visual and auditory environments.

The report details progress in several key areas:

1. Nonlinear Center-Surround Interactions: New experimental measurements have been conducted to compare the responses of retinal ganglion cells to artificial stimuli versus natural images. This research aims to provide insights into the nonlinear processing capabilities of retinal cells.

2. Hierarchical Feature Learning Network: The team has made advancements in developing a biologically inspired hierarchical feature learning network. This includes new results related to learning and detecting junction features, which are essential for understanding visual scenes.

3. Visual Attentional Model: The project has further developed a visual attentional model that incorporates learned top-down influences to predict human attentional shifts. Experimental results are included to demonstrate the effectiveness of this model.

4. Auditory Processing: The report also discusses the adaptation of biologically inspired feature extraction and normalization operations from the visual attentional model to detect prominent syllables and words in speech, highlighting the interdisciplinary nature of the research.

The document emphasizes the importance of understanding the human visual system, particularly the retina, as it is the most effective image-understanding device known. The research aims to bridge the gap between artificial stimuli used in current models and the complexities of natural images, ultimately contributing to advancements in image processing technologies.

In summary, this progress report highlights significant strides in understanding and modeling sensory processing, with implications for both visual and auditory systems, aiming to enhance the technical state-of-the-art in these fields through a biologically inspired approach.