----------------------- Page 1-----------------------

2 Foveated Vision Sensor

and Image Processing – A Review



1 2

Mohammed Yeasin , Rajeev Sharma



1. Department of Electrical and Computer Engineering, University of

Memphis, TN 38152-3180

Email: myeasin@memphis.edu



2. Department of Computer Science and Engineering The Pennsyl-

vania State University, University Park, PA-16802



Abstract. The term foveated vision refers to sensor architectures based on smooth

variation of resolution across the visual field, like that of the human visual system.

The foveated vision, however, is usually treated concurrently with the eye motor

system, where fovea focuses on regions of interest (ROI). Such visual sensors

expected to have wide range of machine vision applications in situations where the

constraint of performance, size, weight, data reduction and cost must be jointly

optimized. Arguably, foveated sensors along with a purposefully planned

acquisition strategy can considerably reduce the complexity of processing and

help in designing superior vision algorithms to extract meaningful information

from visual data. Hence, understanding foveated vision sensors is critical for

designing a better machine vision algorithm and understanding biological vision

system.



This chapter will review the state-of-the-art of the retino-cortical (foveated)

mapping models and sensor implementations based on these models. Despite

some notable advantages foveated sensors have not been widely used due to the

lack of elegant image processing tools. Traditional image processing algorithms

are inadequate when applied directly to a space-variant image representation. A

careful design of low level image processing operators (both the spatial and

frequency domain) can offer a meaningful solution to the above mentioned

problems. The utility of such approach was explefied through the computation of

optical flow on log-mapped images.



Key words Foveated vision, Retino-cortical mapping, Optical flow, Stereo

disparity, Conformal mapping, and Chirp transform.
....