Theses and Dissertations from UMD
Permanent URI for this communityhttp://hdl.handle.net/1903/2
New submissions to the thesis/dissertation collections are added automatically as they are received from the Graduate School. Currently, the Graduate School deposits all theses and dissertations from a given semester after the official graduation date. This means that there may be up to a 4 month delay in the appearance of a give thesis/dissertation in DRUM
More information is available at Theses and Dissertations at University of Maryland Libraries.
Browse
5 results
Search Results
Item THE ACOUSTIC QUALITIES THAT INFLUENCE AUDITORY OBJECT AND EVENT RECOGNITION(2019) Ogg, Mattson Wallace; Slevc, L. Robert; Neuroscience and Cognitive Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Throughout the course of a given day, human listeners encounter an immense variety of sounds in their environment. These are quickly transformed into mental representations of objects and events in the world, which guide more complex cognitive processes and behaviors. Through five experiments in this dissertation, I investigated the rapid formation of auditory object and event representations (i.e., shortly after sound onset) with a particular focus on understanding what acoustic information the auditory system uses to support this recognition process. The first three experiments analyzed behavioral (dissimilarity ratings in Experiment 1; duration-gated identification in Experiment 2) and neural (MEG decoding in Experiment 3) responses to a diverse array of natural sound recordings as a function of the acoustic qualities of the stimuli and their temporal development alongside participants’ concurrently developing responses. The findings from these studies highlight the importance of acoustic qualities related to noisiness, spectral envelope, spectrotemporal change over time, and change in fundamental frequency over time for sound recognition. Two additional studies further tested these results via syntheszied stimuli that explicitly manipulated these acoustic cues, interspersed among a new set of natural sounds. Findings from these acoustic manipulations as well as replications of my previous findings (with new stimuli and tasks) again revealed the importance of aperiodicity, spectral envelope, spectral variability and fundamental frequency in sound-category representations. Moreover, analyses of the synthesized stimuli suggested that aperiodicity is a particularly robust cue for some categories and that speech is difficult to characterize acoustically, at least based on this set of acoustic dimensions and synthesis approach. While the study of the perception of these acoustic cues has a long history, a fuller understanding of how these qualities contribute to natural auditory object recognition in humans has been difficult to glean. This is in part because behaviorally important categories of sound (studied together in this work) have previously been studied in isolation. By bringing these literatures together over these five experiments, this dissertation begins to outline a feature space that encapsulates many different behaviorally relevant sounds with dimensions related to aperiodicity, spectral envelope, spectral variability and fundamental frequency.Item Context Driven Scene Understanding(2015) Chen, Xi; Davis, Larry S; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Understanding objects in complex scenes is a fundamental and challenging problem in computer vision. Given an image, we would like to answer the questions of whether there is an object of a particular category in the image, where is it, and if possible, locate it with a bounding box or pixel-wise labels. In this dissertation, we present context driven approaches leveraging relationships between objects in the scene to improve both the accuracy and efficiency of scene understanding. In the first part, we describe an approach to jointly solve the segmentation and recognition problem using a multiple segmentation framework with context. Our approach formulates a cost function based on contextual information in conjunction with appearance matching. This relaxed cost function formulation is minimized using an efficient quadratic programming solver and an approximate solution is obtained by discretizing the relaxed solution. Our approach improves labeling performance compared to other segmentation based recognition approaches. Secondly, we introduce a new problem called object co-labeling where the goal is to jointly annotate multiple images of the same scene which do not have temporal consistency. We present an adaptive framework for joint segmentation and recognition to solve this problem. We propose an objective function that considers not only appearance but also appearance and context consistency across images of the scene. A relaxed form of the cost function is minimized using an efficient quadratic programming solver. Our approach improves labeling performance compared to labeling each image individually. We also show the application of our co-labeling framework to other recognition problems such as label propagation in videos and object recognition in similar scenes. In the third part, we propose a novel general strategy for simultaneous object detection and segmentation. Instead of passively evaluating all object detectors at all possible locations in an image, we develop a divide-and-conquer approach by actively and sequentially evaluating contextual cues related to the query based on the scene and previous evaluations---like playing a ``20 Questions'' game---to decide where to search for the object. Such questions are dynamically selected based on the query, the scene and current observed responses given by object detectors and classifiers. We first present an efficient object search policy based on information gain of asking a question. We formulate the policy in a probabilistic framework that integrates current information and observation to update the model and determine the next most informative action to take next. We further enrich the power and generalization capacity of the Twenty Questions strategy by learning the Twenty Questions policy driven by data. We formulate the problem as a Markov Decision Process and learn a search policy by imitation learning.Item DOMAIN ADAPTIVE OBJECT RECOGNITION AND DETECTION(2013) Mirrashed, Fatemeh; Davis, Larry S; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Discriminative learning algorithms rely on the assumption that training and test data are drawn from the same marginal probability distribution. In real world applications, however, this assumption is often violated and results in a significant performance drop. We often have sufficient labeled training data from single or multiple "source" domains but wish to learn a classifier which performs well on a "target" domain with a different distribution and no labeled training data. In visual object detection, for example, where the goal is to locate the objects of interest in a given image, it may be infeasible to collect training data to model the enormous variety of possible combinations of pose, background, resolution, and lighting conditions affecting object appearance. Thus, we generally expect to encounter instances or domains at test time for which we have seen little or no training data. To this end, we first propose a framework for domain adaptive object recognition and detection using Transfer Component Analysis, an unsupervised domain adaptation and dimensionality reduction technique. The idea is to obtain a transformation in feature space to a latent subspace that reduces the distance between the source and target data distributions. We evaluate the effectiveness of this approach for vehicle detection using video frames from 50 different surveillance cameras. Next, we explore the problem of extreme class imbalance present when performing fully unsupervised domain adaptation for object detection. The main challenge arises from the fact that images in unconstrained settings are mostly occupied by the background (negative class). Therefore, random sampling will not be effective in obtaining a sufficient number of positive samples from the target domain, which is required by any adaptation method. We propose a variation of co-learning technique that automatically constructs a more balanced set of samples from the target domain. We compare the performance of our technique with other approaches such as unbiased learning from multiple datasets and self-learning. Finally, we propose a novel approach for unsupervised domain adaptation. Our method learns a set of binary attributes for classification that captures the structural information of the data distribution in the target domain itself. The key insight is finding attributes that are discriminative across categories and predictable across domains. We formulate our optimization problem to learn these attributes and the classifier jointly. We evaluate the performance of our method on a wide range of tasks including cross-domain object recognition and sentiment analysis on textual data both in inductive and transductive settings. We achieve a performance that significantly exceeds the state-of-the-art results on standard benchmarks. In many cases we reach the same-domain performance, the upper bound, in unsupervised domain adaptation scenarios.Item Model-driven and Data-driven Approaches for some Object Recognition Problems(2011) Gopalan, Raghuraman; Chellappa, Rama; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Recognizing objects from images and videos has been a long standing problem in computer vision. The recent surge in the prevalence of visual cameras has given rise to two main challenges where, (i) it is important to understand different sources of object variations in more unconstrained scenarios, and (ii) rather than describing an object in isolation, efficient learning methods for modeling object-scene `contextual' relations are required to resolve visual ambiguities. This dissertation addresses some aspects of these challenges, and consists of two parts. First part of the work focuses on obtaining object descriptors that are largely preserved across certain sources of variations, by utilizing models for image formation and local image features. Given a single instance of an object, we investigate the following three problems. (i) Representing a 2D projection of a 3D non-planar shape invariant to articulations, when there are no self-occlusions. We propose an articulation invariant distance that is preserved across piece-wise affine transformations of a non-rigid object `parts', under a weak perspective imaging model, and then obtain a shape context-like descriptor to perform recognition; (ii) Understanding the space of `arbitrary' blurred images of an object, by representing an unknown blur kernel of a known maximum size using a complete set of orthonormal basis functions spanning that space, and showing that subspaces resulting from convolving a clean object and its blurred versions with these basis functions are equal under some assumptions. We then view the invariant subspaces as points on a Grassmann manifold, and use statistical tools that account for the underlying non-Euclidean nature of the space of these invariants to perform recognition across blur; (iii) Analyzing the robustness of local feature descriptors to different illumination conditions. We perform an empirical study of these descriptors for the problem of face recognition under lighting change, and show that the direction of image gradient largely preserves object properties across varying lighting conditions. The second part of the dissertation utilizes information conveyed by large quantity of data to learn contextual information shared by an object (or an entity) with its surroundings. (i) We first consider a supervised two-class problem of detecting lane markings from road video sequences, where we learn relevant feature-level contextual information through a machine learning algorithm based on boosting. We then focus on unsupervised object classification scenarios where, (ii) we perform clustering using maximum margin principles, by deriving some basic properties on the affinity of `a pair of points' belonging to the same cluster using the information conveyed by `all' points in the system, and (iii) then consider correspondence-free adaptation of statistical classifiers across domain shifting transformations, by generating meaningful `intermediate domains' that incrementally convey potential information about the domain change.Item Algorithmic issues in visual object recognition(2009) Hussein, Mohamed Elsayed Ahmed; Davis, Larry; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)This thesis is divided into two parts covering two aspects of research in the area of visual object recognition. Part I is about human detection in still images. Human detection is a challenging computer vision task due to the wide variability in human visual appearances and body poses. In this part, we present several enhancements to human detection algorithms. First, we present an extension to the integral images framework to allow for constant time computation of non-uniformly weighted summations over rectangular regions using a bundle of integral images. Such computational element is commonly used in constructing gradient-based feature descriptors, which are the most successful in shape-based human detection. Second, we introduce deformable features as an alternative to the conventional static features used in classifiers based on boosted ensembles. Deformable features can enhance the accuracy of human detection by adapting to pose changes that can be described as translations of body features. Third, we present a comprehensive evaluation framework for cascade-based human detectors. The presented framework facilitates comparison between cascade-based detection algorithms, provides a confidence measure for result, and deploys a practical evaluation scenario. Part II explores the possibilities of enhancing the speed of core algorithms used in visual object recognition using the computing capabilities of Graphics Processing Units (GPUs). First, we present an implementation of Graph Cut on GPUs, which achieves up to 4x speedup against compared to a CPU implementation. The Graph Cut algorithm has many applications related to visual object recognition such as segmentation and 3D point matching. Second, we present an efficient sparse approximation of kernel matrices for GPUs that can significantly speed up kernel based learning algorithms, which are widely used in object detection and recognition. We present an implementation of the Affinity Propagation clustering algorithm based on this representation, which is about 6 times faster than another GPU implementation based on a conventional sparse matrix representation.