Hierarchical feature extraction using a self-organised retinal receptive field sampling tessellation

Balasuriya, L.S. and Siebert, J.P. (2006) Hierarchical feature extraction using a self-organised retinal receptive field sampling tessellation. Neural Information Processing: Letters and Reviews, 10(4-6),

Full text not currently available from Enlighten.

Abstract

This paper examines the problem of hierarchical processing of visual information extracted with a layer of pseudo-randomly tessellated retinal receptive fields. Afferents from the retinal neural layer were processed by a cortical neuron layer resulting in a hierarchy of feature extraction operations similar to that found in biological vision systems. The retinal tessellation was obtained by self-organization such that retinal neuron receptive field locations were allocated across the system’s field-of-view in a space-variant manner. The retinal tessellation seamlessly merged from a central dense uniform fovea region to a sparser space-variant surrounding periphery. The neural system therefore extracted high acuity visual information from the central foveal region and progressively coarser information from a space-variant increasingly coarse periphery. The paper addresses the issues of i) generating a retinal tessellation with a uniform fovea that seamlessly merges into a space-variant periphery, ii) sampling visual information contained in a digital image with pseudo-randomly positioned retinal neuron receptive fields, iii) performing hierarchical feature extraction on pseudo-randomly spaced visual information, iv) multi-resolution feature extraction using self-organized retinae, v) the targeting of the space-variant machinery by generating saccades based on bottom-up attention. Keywords — Space-variant vision, artificial retina, self-organization 1.

Item Type:Articles (Letter)
Status:Published
Refereed:Yes
Glasgow Author(s) Enlighten ID:Siebert, Dr Jan
Authors: Balasuriya, L.S., and Siebert, J.P.
College/School:College of Science and Engineering > School of Computing Science
Journal Name:Neural Information Processing: Letters and Reviews

University Staff: Request a correction | Enlighten Editors: Update this record