Hue Segmentation, Color Circuitry, and the Mantis Shrimp

Author: Perez, Frank Allen

Year: 1995

Degree: Dissertation (Ph.D.)

Advisors: Antonsson, Erik K.; Koch, Christof

Committee Members: Koch, Christof; Antonsson, Erik K.; Allman, John Morgan; Barr, Alan H.; Culick, Fred E. C.; Mead, Carver; Perona, Pietro

Option: Mechanical Engineering; Computer Science

DOI: 10.7907/zev6-6330

Abstract

The focus of this thesis is on engineering effective color vision subsystems for object segmentation based on hue and discontinuities in hue. Hue, which encodes color, is a psychophysical scalar variable defined on the ring [0, [pi]]. A computational theory justifying the use of hue for distinguishing material differences is established and novel algorithms are developed to detect its discontinuity. Although the focus of this thesis has been on anthropocentrically based trichromatic systems, some effort is placed in exploring the multi-dimensional spaces of more than three primary colors. A hypothetical explanation is proposed of the 11 spectral sensors of the Mantis shrimp, which performs functionally as a biological spectrum analyzer. The importance of the opponency calculation is emphasized. Its role in optimum filter design offers hyperacuity in the spectral domain.

This work encompasses spectral filter design, color space evaluation, computer vision algorithm development, and hardware implementation in custom analog VLSI circuitry. A one pixel (zero dimension) "intensity/normalized-color/hue" sensor is built based on a trichromatic system. For the basic analog circuit element required in hue segmentation, the current-fuse is developed. Guidelines for building higher dimensional sensors in both spatial and spectral domains are presented. Practical offshoots from this research range from color quality sensors for inexpensive printing to vision systems for robotics and autonomous vehicles.

Files