NEURAL BASIS OF SHAPE FROM TEXTURE

State College Of Optometry, New York NY

Investigators

Linked publications & trials

Abstract

Description: Human vision I uses several types of monocular cues such as shading, texture and motion to extract 3D shape from 2D images. This project aims to provide an account of shape perception from texture cues exclusively in terms of feasible neural operations performed by realistic cortical neurons. The proposed work builds on the PIs' recent results showing that observers perceive the true 3D shape of a textured surface only when the image contains visible energy along projected lines of maximum curvature of the surface. Local spectral analyses of convex, concave and slanted portions of corrugated surfaces reveal that contours along projected lines of principal surface curvature show characteristic patterns of orientation and frequency modulations that are systematically related to the 3D shape of the surface. The Pis now seek to develop a model that relates the outputs of V1 neurons to perceived shape from texture cues. The front end of the model will consist of arrays of frequency and orientation-selective neurons. Template patterns of frequency and orientation modulations will be extracted by later cells from set of iso-frequency and iso-orientation outputs. The computational modeling will require new uses of the mathematics of vector fields. Properties of neurons at each level will be tested in single-cell recordings from awake behaving monkeys in a separate collaboration, thus providing a direct link between computation and physiology. To test and extend the model, psychophysical experiments will use patterns with generically occurring orientation distractors, contrast defined contours and naturally occurring textures. The project will provide rules for classifying natural textures as effective or ineffective at conveying shape which will be useful in computer simulations of realistic scenes.