Visual object recognition is a fundamental capability of our visual system - it allows us to identify our friends, recognize abstract symbols such as words, leading to written communication, and provides a range of fundamental survival skills, e.g. which fruit is good to eat? Object recognition is complex, with neurons in the later stages of the brain’s visual pathways selective for high-level representations, e.g. hands. It is logical to have a set of processing stages that leads to highly selective object selectivity but what makes this process computationally challenging is the need to build invariance into the process. That is, we should recognize objects regardless of their size, orientation, luminance or position. It may even be beneficial to simplify objects for easier identification, e.g. cartoonists often draw only three fingers on hands, but most people don’t notice and this practice does not reduce our ability to recognise them as hands. Therefore, to achieve invariance, the visual system must combine fine selectivity for particular features of relevance, with insensitivity to irrelevant features. In a world where “deep convolutional neural networks” have achieved near-human levels of performance in object recognition using massive computing power, it is timely to use new techniques to uncover the richness of the brain’s capacity to achieve the same results with real neural networks. To this end, my talk will outline our latest research on the mechanisms of object recognition in the first few stages of the visual pathway. We have revealed that even in the primary visual cortex, just a few synapses away from the photoreceptors, invariance is already built into the processing mechanisms for multiple object features, thus laying the foundations for complete object recognition at higher levels.
Professor Michael Ibbotson carried out his BSc and PhD in the field of neuroscience at Queen Mary, University of London. He did post-doctoral research at the Australian National University, Canberra, working with Professor Srinivasan and later held a prestigious Australian Research Council Research Fellowship with Professor Mark. At ANU, he became Chair of Faculty at the Research School of Biological Sciences, Head of the Visual Sciences Group, and one of the Associate Directors of the ARC Centre of Excellence in Vision Science. In 2011 he became the Director of the National Vision Research Institute of Australia, based in Melbourne. He is a node leader in the Centre of Excellence for Integrative Brain Function (2014-2021) and was a chief investigator in Bionic Vision Australia (2009-2014). His research interests are focused on how natural visual systems see. This interest has led to a range of experimental approaches that include neurophysiology, neural imaging, eye movement recording, perceptual analysis and computer modelling. Several themes have formed the basis of these studies. Firstly, discovering the physiological and anatomical structures that process visual information, most notably the primary visual cortex, and, secondly, studying how these visual pathways influence behaviour and perception. The third major theme is to utilise this knowledge to develop prosthetic vision devices capable of returning functional sight to the profoundly blind. These prosthetic devices extract images from digital cameras and send highly processed versions of the signals via electrical stimulation to the retina (bionic eye).