The representation of contour shape is an essential component of object recognition, but the cortical mechanisms underlying it are incompletely understood, leaving it a fundamental open question in neuroscience.  Such an understanding would be useful theoretically as well as in developing computer vision and Brain-Computer Interface applications.  We ask two fundamental questions: “How is contour shape represented in cortex and how can neural models and computer vision algorithms more closely approximate this?”

We have created a population of V4-like cells – responsive to a particular local contour conformation located at a specific position on an object’s boundary – and have demonstrated high recognition accuracies classifying handwritten digits in the MNIST database and objects in the MPEG-7 Shape Silhouette database.  We have analyzed the relative contributions of various feature sensitivities to recognition accuracy and robustness to noise, with local curvature the most informative for shape recognition.  We have also shown that top-down inputs can improve classification accuracy by re-weighting the contributions of intermediate-level units.

Using a wide variety of techniques and parameter values, we have created a population of IT-like cells, which integrate specific information about the 2-D boundary shapes of multiple contour fragments.  We have evaluated performance on a set of real images as a function of the V4 cell inputs.  We have determined the sub-populations of cells that present clear preferential responses to specific categories, at the exclusion of others, and have illustrated the fact that different images may activate the IT-like cells differently, yet produce similar total responses – necessary to achieve the robust discriminatory power of our network.  We have classified based upon cell population response and we have concluded that only a small number of V4 / IT cells may be necessary for image recognition at this level.  Finally, we have constructed and validated a biologically-plausible Izhikevich model cell network of V4, IT and Inhibitory cells and have demonstrated an enhanced response in IT, correlated with recognition, via gamma synchronization in V4.

Our results support the hypothesis that the response properties of V4 and IT cells, and in particular their sensitivities to curvatures and contour positions, function as robust shape descriptors and are useful for object recognition precisely because they, like our faithful computer models of them, facilitate shape representation and categorization by extracting features that correlate with global shape.  We have established a connection between a computer model of a recognition system and known biological phenomena.