Texture segregation by visual cortex: Perceptual grouping, attention, and learning

This thesis develops the dARTEX neural model of how interactions in the visual cortex may learn and recognize object texture and form boundaries. Global object form and local surface texture are important cues for identification and recognition of visual objects, and the visual cortex plays an active role in their processing. The dARTEX model proposes how feedforward and feedback interactions between different layers of visual cortical areas V1, V2, and V4, and between V1 and the Lateral Geniculate Nucleus of the thalamus, may realize five processes: region-based texture classification, contour-based boundary grouping, surface filling-in, object attention, and spatial attention. These processes interact to learn texture prototypes, which in turn help to generate better texture boundaries and figural shapes. dARTEX clarifies the role of top-down attention in learning, recognition, and segmentation of textured objects and how both boundaries and surfaces may regulate object and spatial attention. dARTEX can also discriminate abutted textures with blurred boundaries and is sensitive to texture boundary attributes like discontinuities in orientation and texture flow curvature as well as to relative orientations of texture elements. The model quantitatively fits a large set of human psychophysical data on orientationbased textures that is difficult to explain using existing models. Object boundary output of the model is also compared to computer vision algorithms using a set of human segmented photographic images. dARTEX classifies local texture and suppresses noise using a multiple-scale oriented filterbank and a distributed Adaptive Resonance Theory (dART) classifier. The matched signal between the bottom-up texture inputs and top-down learned texture categories is utilized by oriented competitive and cooperative grouping processes to generate texture boundaries 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, improved texture classification performance, and efficient self-supervised learning of texture within attended surface regions. Improvement in classification performance and self-supervised texture learning afforded by the top-down surface-based attention is quantified using comparative performance benchmarks. Benchmark classification rate on Brodatz (1966) images varies from 95.1% to 98.6% with attention, and from 90.6% to 93.2% without attention.