The Roles of Endstopped and Curvature Tuned Computations in a Hierarchical Representation of 2D Shape

摘要: That shape is important for perception has been known almost a thousand years (thanks to Alhazen in 1083) and subject of study ever since by scientists phylosophers (such as Descartes, Helmholtz or the Gestalt psychologists). Shapes are object descriptors. If there was any remote doubt regarding importance shape, recent experiments have shown that intermediate areas primate visual cortex such V2, V4 TEO involved analyzing features corners curvatures. The brain appears perform wide variety complex tasks means simple operations. These operations applied across several layers neurons, representing increasingly complex, abstract processing stages. Recently, new models attempted emulate human system. However, role representations their not adequately studied computational modeling. This paper proposes model shape-selective neurons whose shape-selectivity achieved through representation previously fully explored. We hypothesize hypercomplex - also endstopped play critical achieve selectivity show how may be modeled integrating endstopping curvature computations. This representational system detection 2-dimensional silhouettes we term 2DSIL provides highly accurate fit with neural data replicates responses from area an average 83% accuracy. successfully test biologically plausible hypothesis on connect early based Gabor Difference Gaussian filters later closer categories without need learning phase most models.