Zhaoping Li, University College London
I will present a theory that primary visual cortex creates a saliency map to guide attention exogenously, and show how this explains and predicts experimental data in physiology and in visual behavior. Implications of this theory will be discussed.
Biography: I obtained my B.S. in Physics in 1984 from Fudan University, Shanghai, and Ph.D. in Physics in 1989 from California Institute of Technology. I was a postdoctoral researcher in Fermi National Laboratory in Batavia, Illinois USA, Institute for Advanced Study in Princeton New Jersey, USA, and Rockefeller University in New York USA. I have been a faculty member in Computer Science in Hong Kong University of Science and Technology, and was a visiting scientist at various academic institutions. In 1998, I helped to found the Gatsby Computational Neuroscience Unit in University College London. Currently, I am a Professor of computational neuroscience in the Department of Computer Science in University College London. My research experience throughout the years ranges from areas in high energy physics to neurophysiology and marine biology, with most experience in understanding the brain functions in vision, olfaction, and in nonlinear neural dynamics. In late 90s and early 2000s, I proposed a theory (which is being extensively tested) that the primary visual cortex in the primate brain creates a saliency map to automatically attract visual attention to salient visual locations. I am the author of Understanding Vision: theory, models, and data , Oxford University Press, 2014.
Abstract: `Motion dazzle’ is the hypothesis that certain types of patterns, such as high contrast stripes and zigzags, can cause misperceptions in the speed and direction perception of moving targets. Motion dazzle is relevant to both ecological questions, including why striped patterning may have evolved in animals such as zebras, and also for camouflage design for human purposes. My work addresses the question of whether motion dazzle exists, and what mechanisms may underlie it if it does. I use an interdisciplinary approach, combining techniques from psychophysics and behavioural ecology in testing human subjects. I present data showing that targets with striped markings are among the hardest for humans to ‘capture’ in a touch screen task, but that these effects may depend upon factors such as target orientation and whether multiple targets are present. I also present psychophysical data that shows that subjects make small direction judgement errors that depend upon the orientation of the striped pattern on a target relative to the direction of motion, suggesting a possible mechanism for motion dazzle effects. Finally, I discuss future directions for my research, including a new project involving large scale data collection using citizen science methods.
Abstract: Why do armies operating in the same environment (e.g. temperate woodland) wear markedly different dress? The primary function of military camouflage is generally understood to be concealment, however the vast diversity of camouflage patterns (over 600 patterns in the past century) suggests additional design factors. One hypothesis is that camouflage patterns can also act as signals of alliance and aiding soldiers to distinguish friend from foe. On the other hand, newly independent states can endorse their identity by issuing distinctive camouflage. In both cases designs must remain constrained to function as adequate concealment. The aim of this talk is to demonstrate how a phylogenetic model can be useful for testing these hypotheses. In order to quantify similarity between patterns, I used methods from computer vision to compare their texture and colour. Camouflage of countries can be represented as phylogenies as temporal information (e.g. when patterns were issued) is readily available. Combining computer vision-derived metrics and phylogenetic analysis, I show how certain “design drifts” can be detected throughout the history of camouflage uniforms.
Everyone knows what cosmologists do: they gaze out into the sky to see the secrets of what’s out there, matching observations with theory to understand how the universe came about. Visual psychophysicists are motivated by a similar sense of wonder; but the universe they want to understand lives inside the biological computer of the human brain. By using carefully designed visual stimuli in controlled laboratory environments, it turns out that this non-invasive research method can provide an incisive view of the visual mechanisms that lie within us, whether we are conscious of them or not! I shall show how psychophysics has been used to see into the very early stages of the visual processing stream, where the ‘atoms’ of vision are very different from the rich visual experience that they deliver when we open our eyes, and where competing theories have been pitched against one another, helping us to understand how we see.
Abstract: What is the best way to represent the three-dimensional world we see on a two-dimensional surface? For several hundred years there was basically one answer to this question: use linear perspective. Linear perspective is a method of mapping rays of light onto an image plane in a way that corresponds to well-understood laws of geometry and optics. It now underpins nearly all existing imaging technology. But even the artists who first developed linear perspective understood that it had severe limitations when it comes to representing the wide angled, binocular vision most humans enjoy. As has been pointed out many times, linear perspectival images viewed under everyday conditions do not replicate our visual experience of the world very well. Over the last two centuries various alternative methods have been proposed, normally by artists, which are claimed to better represent human visual perception. These are usually based on subjective judgments about how things appear to human observers rather the objective behaviour of light. Such methods, while interesting historically, were of limited utility to image-makers as they were even more complicated to apply than the rules of linear perspective. However, the current state of computer graphics allows us to manipulate image data in powerful ways, and to design systems that incorporate some of the subjective properties of vision that it was not previously practical to model. In this talk I will outline the approach my lab is taking to designing new technologies for representing human vision that are an improvement on what can currently be achieved with existing linear perspective-based technology. This will include some discussion of the art historical background to our work, the scientific data we have gathered, and demonstrations of some the technological solutions we have developed.
Abstract: In this talk, we combine the methods from probability theory and stochastic geometry to put forward new solutions to the multiple object detection and tracking problem in high resolution remotely sensed image sequences. First, we present a spatial marked point process model to detect a pre-defined class of objects based on their visual and geometric characteristics. Then, we extend this model to the temporal domain and create a framework based on spatio-temporal marked point process models to jointly detect and track multiple objects in image sequences. We propose the use of simple parametric shapes to describe the appearance of these objects. We build new, dedicated energy based models consisting of several terms that take into account both the image evidence and physical constraints such as object dynamics, track persistence and mutual exclusion. We construct a suitable optimization scheme that allows us to find strong local minima of the proposed highly non-convex energy. As the simulation of such models comes with a high computational cost, we turn our attention to the recent filter implementations for multiple objects tracking, which are known to be less computationally expensive. We propose a hybrid sampler by combining the Kalman filter with the standard Reversible Jump MCMC. High performance computing techniques are also used to increase the computational efficiency of our method. We provide an analysis of the proposed framework. This analysis yields a very good detection and tracking performance at the price of an increased complexity of the models. Tests have been conducted both on high resolution satellite and microscopy image sequences.