http://hdl.handle.net/1721.1/7776 http://hdl.handle.net/1721.1/49739 Perceptually inspired image estimation and enhancement Li, Yuanzhen, Ph. D. Massachusetts Institute of Technology In this thesis, we present three image estimation and enhancement algorithms inspired by human vision. In the first part of the thesis, we propose an algorithm for mapping one image to another based on the statistics of a training set. Many vision problems can be cast as image mapping problems, such as, estimating reflectance from luminance, estimating shape from shading, separating signal and noise, etc. Such problems are typically under-constrained, and yet humans are remarkably good at solving them. Classic computational theories about the ability of the human visual system to solve such under-constrained problems attribute this feat to the use of some intuitive regularities of the world, e.g., surfaces tend to be piecewise constant. In recent years, there has been considerable interest in deriving more sophisticated statistical constraints from natural images, but because of the high-dimensional nature of images, representing and utilizing the learned models remains a challenge. Our techniques produce models that are very easy to store and to query. We show these techniques to be effective for a number of applications: removing noise from images, estimating a sharp image from a blurry one, decomposing an image into reflectance and illumination, and interpreting lightness illusions. In the second part of the thesis, we present an algorithm for compressing the dynamic range of an image while retaining important visual detail. The human visual system confronts a serious challenge with dynamic range, in that the physical world has an extremely high dynamic range, while neurons have low dynamic ranges. (cont.) The human visual system performs dynamic range compression by applying automatic gain control, in both the retina and the visual cortex. Taking inspiration from that, we designed techniques that involve multi-scale subband transforms and smooth gain control on subband coefficients, and resemble the contrast gain control mechanism in the visual cortex. We show our techniques to be successful in producing dynamic-range-compressed images without compromising the visibility of detail or introducing artifacts. We also show that the techniques can be adapted for the related problem of "companding", in which a high dynamic range image is converted to a low dynamic range image and saved using fewer bits, and later expanded back to high dynamic range with minimal loss of visual quality. In the third part of the thesis, we propose a technique that enables a user to easily localize image and video editing by drawing a small number of rough scribbles. Image segmentation, usually treated as an unsupervised clustering problem, is extremely difficult to solve. With a minimal degree of user supervision, however, we are able to generate selection masks with good quality. Our technique learns a classifier using the user-scribbled pixels as training examples, and uses the classifier to classify the rest of the pixels into distinct classes. It then uses the classification results as per-pixel data terms, combines them with a smoothness term that respects color discontinuities, and generates better results than state-of-art algorithms for interactive segmentation. Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2009. Includes bibliographical references (p. 137-144). Wed, 29 Oct 2008 22:58:59 GMT http://hdl.handle.net/1721.1/49738 Deficient experience-dependent plasticity in the visual cortex of Arc null mice McCurry, Cortina (Cortina Luann) Within the visual cortex a vast assortment of molecules work in concert to sharpen and refine neuronal circuits throughout development. With the advent of genetic mouse models it is now possible to probe the individual contributions of single molecules implicated in this process. The Arc (activity-regulated cytoskeletal associated) gene is an effector immediate early gene that has been suggested to play a critical role in synaptic plasticity. The goal of this thesis is to understand the workings of Arc within the visual cortex. Specifically, we ask how genetic deletion of Arc influences plasticity, and how visual response properties differ between cells types containing, and not containing Arc. To elucidate a role for Arc in visual cortical plasticity we took advantage of knockin mice expressing GFP in place of Arc protein (referred to as KO mice for simplicity). We combined intrinsic signal imaging, visually evoked potentials, and two-photon in vivo calcium imaging to assess plasticity in juvenile and adult wild-type (WT), heterozygote, and KO mice. We find that plasticity is disrupted in the visual cortex of Arc KO mice in the absence of obvious deficits at the level of basal response properties. In addition, this work has revealed that: 1) Arc is necessary for the establishment of normal ocular dominance during development and critical for deprived eye depression in the visual cortex of juvenile animals 2) Loss of Arc impairs AMPA receptor internalization in visual cortex- a necessary requirement for synaptic weakening after lid suture. (cont.) 3) Open eye potentiation fails to occur after extended deprivation in the absence of Arc 4) Arc is required for stimulus response potentiation in juvenile animals. 5) Arc is not required for the synaptic scaling up of response suggesting a specific role in Hebbian plasticity. 6) Single cell analysis within the binocular zone of Arc-GFP homozygotes reveals that the distribution of Arc lacking GFP-positive cells does not display a contralateral-bias as compared to controls, and the majority of Arc-lacking GFP-positive cells receive equal input from each eye, suggesting that Arc is critical for synaptic weakening during development. Together, these experiments illustrate the essential role for Arc in experience-dependent plasticity within the visual system. Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2009. Vita. Includes bibliographical references. Wed, 29 Oct 2008 22:58:59 GMT http://hdl.handle.net/1721.1/47892 Natively probabilistic computation Mansinghka, Vikash Kumar I introduce a new set of natively probabilistic computing abstractions, including probabilistic generalizations of Boolean circuits, backtracking search and pure Lisp. I show how these tools let one compactly specify probabilistic generative models, generalize and parallelize widely used sampling algorithms like rejection sampling and Markov chain Monte Carlo, and solve difficult Bayesian inference problems. I first introduce Church, a probabilistic programming language for describing probabilistic generative processes that induce distributions, which generalizes Lisp, a language for describing deterministic procedures that induce functions. I highlight the ways randomness meshes with the reflectiveness of Lisp to support the representation of structured, uncertain knowledge, including nonparametric Bayesian models from the current literature, programs for decision making under uncertainty, and programs that learn very simple programs from data. I then introduce systematic stochastic search, a recursive algorithm for exact and approximate sampling that generalizes a popular form of backtracking search to the broader setting of stochastic simulation and recovers widely used particle filters as a special case. I use it to solve probabilistic reasoning problems from statistical physics, causal reasoning and stereo vision. Finally, I introduce stochastic digital circuits that model the probability algebra just as traditional Boolean circuits model the Boolean algebra. (cont.) I show how these circuits can be used to build massively parallel, fault-tolerant machines for sampling and allow one to efficiently run Markov chain Monte Carlo methods on models with hundreds of thousands of variables in real time. I emphasize the ways in which these ideas fit together into a coherent software and hardware stack for natively probabilistic computing, organized around distributions and samplers rather than deterministic functions. I argue that by building uncertainty and randomness into the foundations of our programming languages and computing machines, we may arrive at ones that are more powerful, flexible and efficient than deterministic designs, and are in better alignment with the needs of computational science, statistics and artificial intelligence. Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2009. Includes bibliographical references (leaves 129-135). Wed, 29 Oct 2008 22:58:59 GMT http://hdl.handle.net/1721.1/47891 The rational child : theories and evidence in prediction, exploration, and explanation Bonawitz, Elizabeth R. (Elizabeth Robbin) In this thesis, rational Bayesian models and the Theory-theory are bridged to explore ways in which children can be described as Bayesian scientists. I investigate what it means for children to take a rational approach to processes that support learning. In particular, I present empirical studies that show children making rational predictions, exploration, and explanations. I test the claim that differences in prior beliefs or changes in the observed evidence should affect these behaviors. The studies presented in this thesis encompass two manipulations: in some conditions, children's prior beliefs are equal, but the patterns of evidence are varied; in other conditions, children observe identical evidence but children's prior beliefs are varied. I incorporate an additional approach in this thesis, testing children within a variety of domains, tapping into their intuitive theories of biological kinds, psychosomatic illness, balance, and physical systems. Chapter One introduces the problem. Chapter Two explores how evidence and children's strong beliefs about biological events and psychosomatic illness influence their forced-choice explanations in a story-book task. Chapter Three presents a training study to further investigate the developmental differences discussed in Chapter Two. Chapter Four looks at how children's strong differential beliefs of balance interact with evidence to affect their predictions, play, explanations, and learning. (cont.) Chapter Five looks at children's exploratory play with a jack-in-the-box, (where children don't have strong, differential beliefs), given different patterns of evidence. Chapter Six investigates children's explanations following theory-neutral evidence about a mechanical toy. Chapter Seven concludes the thesis. The following chapters will suggest that frameworks combining evidence and theories capture children's causal learning about the world. Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Brain and Cognitive Sciences, 2009. Includes bibliographical references (p. 122-133). Wed, 29 Oct 2008 22:58:59 GMT