Purchase this article with an account.
Rushi P. Bhatt, Gail A. Carpenter, Stephen Grossberg; Learning and recognition of textured objects. Journal of Vision 2005;5(8):868. doi: https://doi.org/10.1167/5.8.868.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
A neural model is proposed of how laminar interactions in the visual cortex may learn and recognize both object form and texture. The model brings together four interacting processes: region-based texture classification, contour-based boundary grouping, surface filling-in, and spatial attention. These processes interact to learn texture prototypes, which in turn generate better texture boundaries, as well as figural shapes. The model can perform discrimination of abutted textures with blurred boundaries (e.g., Gurnsey & Laundry 1992, Canad. J. Psych.) and shows sensitivity to texture boundaries including those due to discontinuities in orientation (e.g., Nothdurft 1992, Percept. and Psychophy.), texture flow curvature (e.g., Ben-Shahar and Zucker 2004, Vis. Res.), and relative orientations of texture boundary and texture elements (e.g., Wolfson and Landy 1995, Vis. Res.). Object boundary output of the model is also benchmarked against the performance of human subjects and some popular computer algorithms using a database of natural images (Martin et al. 2001, ICCV). The model achieves near-perfect classification performance on a set of texture images chosen from the Brodatz micro-texture album (Brodatz , 1966). In the model, texture is categorized using a multi-scale oriented filter-bank and a distributed Adaptive Resonance Theory (dART) classifier which together classify textures and suppress noise. The matched signal between the bottom-up inputs and top-down learned texture categories is further processed using oriented competitive and cooperative grouping processes to generate texture boundary groupings that control surface filling-in and spatial attention. Top-down modulatory attentional feedback from boundary and surface representations to early filtering stages results in enhanced texture boundaries and more efficient learning of texture classification within attended surface regions.
This PDF is available to Subscribers Only