VASC Seminar Announcement
        =========================
Speaker: Vladimir Brajovic
Affiliation: Carnegie Mellon University, Robotics Institute
E-mail: brajovic@cs.cmu.edu
Date: Monday, November 25th
Time: 3:30pm
Place: NSH 1507

Title: Reflectance Perception: A Neuromorphic Approach

Abstract:
Natural scenes produce images with a dynamic range of more than 1:1,000,000.
The primary reason for such a wide dynamic range is the illumination field
that could vary as much as 1: 100,000 across a scene. The scene reflectance
exhibits variations of only about 1:100. In high dynamic range scenes,
humans routinely perceive details in both bright regions and deep shadows.
This is because human vision is mostly sensitive to the scene
reflectance and not the illumination field.

In this talk I will introduce a possible model for local adaptation that may
be responsible for reflectance perception in humans. The model uses a single
image to estimate the illumination field. The illumination field is then
compensated for by compressing the dynamic range. It is easily solved
mathematically, leading to an elegant implementation in neuromorphic
circuits using a modified cellular neural network.

Illumination induced variations are one of the main difficulties in vision
algorithms. I will show a number of interesting results in which the drastic
influence of the illumination is removed allowing high dynamic range images
to be displayed on low dynamic range screens. In addition, we quantify the
effect of applying the algorithm by showing performance improvements of
standard face recognition algorithms on databases with challenging
illumination conditions.