Master of Science, Miami University, 2019, Computer Science and Software Engineering

Having high resolution Infra-Red (IR) imagery in cluttered environment of battlespace is crucial for capturing intelligence in search and target acquisition tasks such as whether or not a vehicle (or any heat source) has been moved or used and in which direction. While 3D graphic simulation of large scenes helps with retrieving information and training analysts, using traditional 3D rendering techniques are not enough, and an additional parameter needs to be solved due to different concept of visibility in IR scenes. In 3D rendering of IR scenes, the problem of what can currently be seen by a participant of the simulation does not just depend on emitted thermal energy from objects, and the visibility also depends on previous scenes as thermal energy is slowly retained and diffused over time. Therefore, time as an additional factor must be included since the aggregation of heat energy in the scene relates to its past. Our solution uses lightmaps for storing energy that reaches surfaces over time. We modify the lightmaps to solve the problem of lightmap parameterization between 3D surfaces and 2D mapping and add an extra ability to let us periodically update only necessary areas based on dynamic aspects of the scene.

Committee: John Femiani Dr. (Advisor); Eric Bachmann Dr. (Committee Member); Vijayalakshmi Ramasamy Dr. (Committee Member) Subjects: Computer Science

Master of Fine Arts, The Ohio State University, 2016, Theatre

Theatre is a liminal environment between performers and a live audience, and between the past, present and future. Theatre practitioners often bring to life old scripts that have graced the stage many times while highlighting the relevance of key themes and motifs in relation to a modern audience. The work of playwright Henrik Ibsen is produced worldwide because of its modern subjects, despite having been written in the late nineteenth century.Under the direction of Lesley Ferris, I designed the scenic environment for Rebecca Lenkiewicz's version of Henrik Ibsen's An Enemy of the People at The Ohio State University. I used a combination of sketches, digital modelling, and a physical white model to communicate my scenic design. By way of reducing, reusing, and recycling, I executed a sustainable scenic environment that complimented the themes of environmental awareness within the play without compromising the aesthetic of the design.

Committee: Brad Steinmetz M.F.A. (Advisor); Mary Tarantino M.F.A (Committee Member); Lesley Ferris PhD (Committee Member) Subjects: Architectural; Architecture; Art History; Design; Environmental Education; Environmental Health; Environmental Management; Environmental Studies; Fine Arts; Gender; Gender Studies; Performing Arts; Scandinavian Studies; Theater; Theater History; Theater Studies; Womens Studies

Master of Fine Arts, The Ohio State University, 2023, Theatre

Performance is used as a means of storytelling, to escape, to reflect, to learn, to celebrate, or to understand other perspectives. The Country Wife by William Wycherley is all of the above. It is a multi-faceted script that Wycherley curated to appeal to the masses. Written in 1675, this Restoration play uses wit and comedy to comment on societal structures such as gender norms, class, rank, and relationships. The narrative is funny, raunchy, clever, pointed, self-reflective, and opens itself up to be consumed at a variety of levels. I designed the scenery for the 2022 production at The Ohio State University. This is the detailed analysis of that process.

Committee: Brad Steinmetz (Advisor); Tom Dugdale (Committee Member); Sarah Neville (Committee Member) Subjects: Design; Fine Arts; Performing Arts; Theater; Theater History

Master of Science, The Ohio State University, 2011, Computer Science and Engineering

We present a complete end-to-end framework to detect and exploit entry and exit regions in video using behavioral models for object trajectories. We first describe how weak tracking data (short and frequently broken tracks) may be utilized to hypothesize entry and exit regions by constructing the weak tracks into a more usable set of "entity" tracks. The entities provide a more reliable set of entry and exit observations which are clustered to produce a set of potential entry and exit regions within a scene. A behavior-based reliability metric is then used to score each potential entry and exit region, and unreliable regions are removed. Using the detected regions, we then present a method to learn scene occlusions and causal relationships between entry-exit pairs. An extension is also presented that allows our entry/exit detection algorithm to detect global entry and exit regions with respect to the viewspace of a pan-tilt-zoom camera. We provide thorough evaluation of our local and viewspace region discovery approaches, including quantitative experiments, and compare our local method to existing approaches. We also provide experimental results for our region exploitation methods (occlusion discovery and entry-exit region relationships), and demonstrate that they may be incorporated to aid in tasks such as tracking and anomaly detection.

Committee: James Davis Prof. (Advisor); Richard Parent Prof. (Committee Member) Subjects: Computer Science

Doctor of Philosophy, The Ohio State University, 2007, Computer and Information Science

A human listener's ability to organize the time-frequency (T-F) energy of the same sound source into a single stream is termed auditory scene analysis (ASA). Computational auditory scene analysis (CASA) seeks to organize sound based on ASA principles. This dissertation presents a systematic effort on sequential organization in CASA. The organization goal is to group T-F segments from the same speaker that are separated in time into a single stream. This dissertation proposes a speaker-model-based sequential organization framework and it shows better grouping performance than feature-based methods. Specifically, a computational objective is derived for sequential grouping in the context of speaker recognition for multi-talker mixtures. This formulation leads to a grouping system that searches for the optimal grouping of separated speech segments. A hypothesis pruning method is then proposed that significantly reduces search space and time while achieving performance close to that of exhaustive search. Evaluations show that the proposed system improves both grouping performance and speech recognition accuracy. The proposed system is then extended to handle multi-talker as well as non-speech intrusions using generic models. The system is further extended to deal with noisy inputs from unknown speakers. It employs a speaker quantization method that extracts generic models from a large speaker space. The resulting grouping performance is only moderately lower than that with known speaker models. In addition, this dissertation presents a systematic effort in robust speaker recognition. A novel usable speech extraction method is proposed that significantly improves recognition performance. A general solution is proposed for speaker recognition under additive-noise conditions. Novel speaker features are derived from auditory filtering, and are used in conjunction with an uncertainty decoder that accounts for mismatch introduced in CASA front-end processing. Evaluations show (open full item for complete abstract)

Committee: DeLiang Wang (Advisor) Subjects: Computer Science

Master of Fine Arts, The Ohio State University, 1991, Theatre

A paper detailing the design concept and execution of the scenery design for the production of Peter Barnes Red Noses as presented under the direction of Guest Director Paoli Lacy by the Department of Theatre at The Ohio State University. The objective was to create a design which emphasized the control instituted by religion over the masses while showing the ability of laughter to exist, and even flourish, in a world filled with pain and suffering. Included are discussions of the production process, design choices and an evaluation of the successes and failures of the realized designs.

Committee: Russell Hastings (Advisor) Subjects:

Doctor of Philosophy (Ph.D.), University of Dayton, 2024, Computer Science

Image is used everywhere since it conveys a story, a fact, or an imagination without any words. Thus, it can substitute the sentences because the human brain can extract the knowledge from images faster than words. However, creating an image from scratch is not only time-consuming, but also a tedious task that requires skills. Creating an image is not a trivial task since it contains rich features and fine-grained details, such as colors, brightness, saturation, luminance, texture, shadow, and so on. Thus, in order to generate an image in less time and without any artistic skills, sketch-to-image synthesis can be used. The reason is that hand sketches are much easier to produce, where only the key structural information is contained. Moreover, it can be drawn without skills and in less time. In fact, since sketches are often simple and rough black and white and sometimes imperfect, converting a sketch into an image is not a trivial problem. Hence, it has attracted the researchers' attention to solve this challenging problem; therefore, much research has been conducted in this field to generate photorealistic images. However, the generated images still suffer from issues, such as the un-naturality, the ambiguity, the distortion, and most importantly, the difficulty in generating images from complex input with multiple objects. Most of these problems are due to converting a sketch into an image directly in one-shot. To this end, in this dissertation, we propose a new framework that divides the problem into sub-problems, leading to generating high-quality photorealistic images even with complicated sketches. Instead of directly mapping the input sketch into an image, we map the sketch into an intermediate result, namely, mask map, through an instance segmentation and semantic segmentation in two levels: background segmentation and foreground segmentation. Background segmentation is formed based on the context of the existing foreground objects. Various natural scenes a (open full item for complete abstract)

Committee: Tam Nguyen (Committee Chair); James Buckley (Committee Member); Luan Nguyen (Committee Member); Ju Shen (Committee Member) Subjects: Artificial Intelligence; Computer Science

Doctor of Philosophy, The Ohio State University, 2024, Civil Engineering

3D scene modeling techniques serve as the bedrocks in the geospatial engineering and computer science, which drives many applications ranging from automated driving, terrain mapping, navigation, virtual, augmented, mixed, and extended reality (for gaming and movie industry etc.). This dissertation presents a fraction of contributions that advances 3D scene modeling to its state of the art, in the aspects of both appearance and geometry modeling. In contrast to the prevailing deep learning methods, as a core contribution, this thesis aims to develop algorithms that follow first principles, where sophisticated physic-based models are introduced alongside with simpler learning and inference tasks. The outcomes of these algorithms yield processes that can consume much larger volume of data for highly accurate reconstructing 3D scenes at a scale without losing methodological generality, which are not possible by contemporary complex-model based deep learning methods. Specifically, the dissertation introduces three novel methodologies that address the challenges of inferring appearance and geometry through physics-based modeling. Firstly, we address the challenge of efficient mesh reconstruction from unstructured point clouds—especially common in large and complex scenes. The proposed solution employs a cutting-edge framework that synergizes a learned virtual view visibility with graph-cut based mesh generation. We introduce a unique three-step deep network that leverages depth completion for visibility prediction in virtual views, and an adaptive visibility weighting in the graph-cut based surface. This hybrid approach enables robust mesh reconstruction, overcoming the limitations of traditional methodologies and showing superior generalization capabilities across various scene types and sizes, including large indoor and outdoor environments. Secondly, we delve into the intricacies of combining multiple 3D mesh models, particularly those obtained through oblique ph (open full item for complete abstract)

Committee: Rongjun Qin (Advisor); Alper Yilmaz (Committee Member); Charles Toth (Committee Member) Subjects: Civil Engineering

PHD, Kent State University, 2023, College of Education, Health and Human Services / School of Health Sciences

Although speech recognition is often experienced as relatively effortless, there are a number of common challenges that can make speech perception more difficult and may greatly impact speech intelligibility (e.g., environmental noise). However, there is some indication that visual cues can be also used to improve speech recognition (Baratchu et al., 2008) — especially when the visual information is congruent with the speech signal (e.g., talking faces; Massaro, 2002). However, it is less clear how noisy visual environments may impact speech perception when the visual signal is not congruous with the speech signal. In fact, adding incongruous visual information will likely detract precious cognitive resources away from the auditory process, making speech perception in noise a more cognitively difficult task. Therefore, the purpose of this dissertation was to examine cognitive processing effort by measuring changes in pupillary response during the processing of speech in noise paired with incongruous visual noise. The primary hypothesis was that noisy visual information would negatively impact the processing of speech in noisy environments and that would result in a greater pupil diameter. To test this I used a common eye-tracking measure (i.e., pupillometry) to assess the cognitive processing effort needed to process speech in the presence of congruent and incongruous visual noise. The results indicated that visual noise recruits cognitive processing effort away from the auditory signal. Results also indicated that different combinations of auditory and visual noise have a significant impact on cognitive processing effort, which led to an increase in pupil dilation response during speech perception.

Committee: JENNIFER ROCHE (Advisor); BRADLEY MORRIS (Committee Member); BRUNA MUSSOI (Committee Member); JOCELYN FOLK (Other) Subjects: Audiology; Cognitive Psychology; Neurosciences

Master of Science (MS), Wright State University, 2022, Computer Science

Education about implicit bias in clinical settings is essential for improving the quality of healthcare for underrepresented groups. Such a learning experience can be delivered in the form of a serious game simulation. WrightLIFE (Lifelike Immersion for Equity) is a project that combines two serious game simulations, with each addressing the group that faces implicit bias. These groups are individuals that identify as LGBTQIA+ and people with autism spectrum disorder (ASD). The project presents healthcare providers with a training tool that puts them in the roles of the patient and a medical specialist and immerses them in social and clinical settings. WrightLIFE games are distributed on both mobile and desktop devices and go through the entire cycle of providing healthcare professionals with experiential learning, which starts with defining the goals of the simulation and ends with collecting feedback. In this thesis work, cross-platform software frameworks like the Unity Engine have been used to develop survey scenes to comprehensively document users' pre- and post-simulation experience and attitudes towards implicit bias. Life course scenes were designed to convey an enhanced user experience that bridges the socio-technical gap between the real and virtual worlds. By applying existing user-experience design methodologies to design the survey scenes and life course scenes, it was possible to create an immersive experiential-learning assessment tool that has the potential to deliver data-driven and targeted learning.

Committee: Ashutosh Shivakumar Ph.D. (Committee Chair); Yong Pei Ph.D. (Committee Co-Chair); Paul J. Hershberger Ph.D. (Committee Member); Thomas Wischgoll Ph.D. (Committee Member) Subjects: Computer Engineering; Computer Science

Master of Science in Electrical Engineering, University of Dayton, 2022, Electrical and Computer Engineering

Visible and short-wave infrared (Vis-SWIR) hyperspectral imaging (HSI) is a powerful technology for detecting and identifying objects of interest in remote sensing scenarios. HSI sensors suffer from pixel-to-pixel response nonuniformity that manifests as a fixed pattern noise (FPN) in collected image data that ultimately has a negative effect on detection performance. FPN is often removed using flat-field calibration procedures; however, residual FPN often persists within the data and the only option to further remove it is to apply algorithmic approaches that exploit properties of the imaged scene within the data. There are a number of such scene-based nonuniformity correction (SBNUC) approaches in the literature that make assumptions about the stochastic distributions of target and background classes within the data, often leading to undesirable artifacts within the corrected HSI images when these assumptions are violated. In this work, a nonuniformity correction algorithm is developed to remove the FPN based upon separating pixels into different zones based upon their spectral and intensity distributions. We then adapt a previous SBNUC algorithm's ratio-based approach to estimate the FPN within each zone. The approach demonstrates the ability to adapt to different backgrounds within a scene while mitigating artifacts often encountered with other techniques. The performance of the algorithm is evaluated and compared to other SBNUC methods.

Committee: Bradley Ratliff (Advisor); Jason Kaufman (Committee Member); Eric Balster (Committee Member) Subjects: Electrical Engineering

Master of Science (MS), Wright State University, 2020, Physiology and Neuroscience

This thesis explores the interaction between emotions and visual perception using large scale spatial environment as the medium of this interaction. Emotion has been documented to have an early effect on scene perception (Olofsson, Nordin, Sequeira, & Polich, 2008). Yet, most popularly-used scene stimuli, such as the IAPS or GAPED stimulus sets often depict salient objects embedded in naturalistic backgrounds, or “events” which contain rich social information, such as human faces or bodies. And thus, while previous studies are instrumental to our understanding of the role that social-emotion plays in visual perception, they do not isolate the effect of emotion from the social effects in order to address the specific role that emotion plays in scene recognition – defined here as the recognition of large-scale spatial environments. To address this question, we examined how early emotional valence and arousal impact scene processing, by conducting an Event-Related Potential (ERP) study using a well-controlled set of scene stimuli that reduced the social factor, by focusing on natural scenes which did not contain human faces or actors. The study comprised of two stages. First, we collected affective ratings of 440 natural scene images selected specifically so they will not contain human faces or bodies. Based on these ratings, we divided our scene stimuli into three distinct categories: pleasant, unpleasant, and neutral. In the second stage, we recorded ERPs from a separate group of participants as they viewed a subset of 270 scenes ranked highest in each of their respective categories. Scenes were presented for 200ms, back-masked using white noise, while participants performed an orthogonal fixation task. We found that emotional valence had significant impact on scene perception in which unpleasant scenes had higher P1, N1 and P2 peaks. However, we studied the relative contribution of emotional effect and low-level visual features using dominance analysis which can co (open full item for complete abstract)

Committee: Assaf Harel Ph.D. (Advisor); Kathrin L. Engisch Ph.D. (Committee Member); Tamera R. Schneider Ph.D. (Committee Member) Subjects: Neurosciences; Physiology

Doctor of Philosophy, The Ohio State University, 2020, Computer Science and Engineering

In daily listening environments, the speech reaching our ears is commonly corrupted by both room reverberation and background noise. These distortions can be detrimental to speech intelligibility and quality, and also pose a serious problem for many speech-related applications, including automatic speech and speaker recognition. The objective of this dissertation is to enhance speech signals distorted by reverberation and noise, to benefit both human communications and human-machine interaction. Different from traditional signal processing approaches, we employ deep learning approaches to perform reverberant-noisy speech enhancement. Our study starts with speech dereverberation without background noise. Reverberation consists of sound wave reflections from various surfaces in an enclosed space. This means the reverberant signal at any time step includes the damped and delayed past signals. To explore such relationships at different time steps, we utilize a self-attention mechanism as a pre-processing module to produce dynamic representations. With these enhanced representations, we propose a temporal convolutional network (TCN) based speech dereverberation algorithm. Systematic evaluations demonstrate the effectiveness of the proposed algorithm in a wide range of reverberant conditions. Then we propose a deep learning based time-frequency (T-F) masking algorithm to address both reverberation and noise. Specifically, a deep neural network (DNN) is trained to estimate the ideal ratio mask (IRM), in which the anechoic-clean speech is considered as the desired signal. The enhanced speech is obtained by applying the estimated mask to the reverberant-noisy speech. Listening tests show that the proposed algorithm can improve speech intelligibility for hearing-impaired (HI) listeners substantially, and also benefit normal-hearing (NH) listeners. Considering the different natures of reverberation and noise, we propose to perform speech enhancement using a two-stage (open full item for complete abstract)

Committee: DeLiang Wang (Advisor); Eric Fosler-Lussier (Committee Member); Eric Healy (Committee Member) Subjects: Computer Science; Engineering

Master of Science (MS), Wright State University, 2020, Physiology and Neuroscience

During scene perception, studies have shown the importance of the global distribution of a scene. Electrophysiological studies have found these global effects concentrated corresponding to the second positive and first negative peaks (P2 and N1, respectively) of the Event-related potential (ERP) during the first 600 ms of scene perception. We sought to understand in Experiment 1, to what extent early responses to scenes were driven by mid-level global information such as the degree of naturalness or openness in a scene image in the absence of specific low-and high-level information (color and semantic object detail). This was done using artificially-generated stimuli controlling for two global scene properties (GSPs) of spatial boundary and naturalness while minimizing color and semantic object information. Significant effects were observed on the P2 and N1 components as well as the P1 component. However, the question of whether scene perception is dominated by global or local factors had yet to be answered leading to Experiment 2. During Experiment 2, for half the trials scenes were presented in an inverted orientation. We found only an orientation interaction approaching significance corresponding to the P1 time course.

Committee: Assaf Harel Ph.D. (Advisor); Sherif M. Elbasiouny Ph.D. (Committee Member); Joe Houpt Ph.D. (Committee Member) Subjects: Physiological Psychology

Doctor of Philosophy, The Ohio State University, 2019, Computer Science and Engineering

Speech is the most important means of human communication. In real environments, speech is often corrupted by acoustic inference, including noise, reverberation and competing speakers. Such interference leads to adverse effects on audition, and degrades the performance of speech applications. Inspired by the principles of human auditory scene analysis (ASA), computational auditory scene analysis (CASA) addresses speech separation in two main steps: segmentation and grouping. With noisy speech decomposed into a matrix of time-frequency (T-F) units, segmentation organizes T-F units into segments, each of which corresponds to a contiguous T-F region and is supposed to originate from the same source. Two types of grouping are then performed. Simultaneous grouping aggregates segments overlapping in time to simultaneous streams. In sequential grouping, simultaneous streams are grouped across time into distinct sources. As a traditional speech separation approach, CASA has been successfully applied in various speech-related tasks. In this dissertation, we revisit conventional CASA methods, and perform related tasks from a deep learning perspective. As an intrinsic characteristic of speech, pitch serves as a primary cue in many CASA systems. A reliable estimate of pitch is important not only for extracting harmonic patterns at a frame level, but also for streaming voiced speech in sequential grouping. Based on the types of interference, we can divide pitch tracking in two categories: single pitch tracking in noise and multi-pitch tracking. Pitch tracking in noise is challenging as the harmonic structure of speech can be severely contaminated. To recover the missing harmonic patterns, we propose to use long short-term memory (LSTM) recurrent neural networks (RNNs) to model sequential dynamics. Two architectures are investigated. The first one is conventional LSTM that utilizes recurrent connections to model temporal dynamics. The second one is two-level time-frequency (open full item for complete abstract)

Committee: DeLiang Wang (Advisor); Eric Fosler-Lussier (Committee Member); Alan Ritter (Committee Member) Subjects: Computer Science; Engineering

Master of Science (M.S.), University of Dayton, 2019, Electrical Engineering

Hyperspectral imaging sensors suffer from pixel-to-pixel response nonuniformity that manifests as fixed pattern noise (FPN) in collected data. FPN is typically removed by application of flat-field calibration procedures and nonuniformity correction algorithms. Despite application of these techniques, some amount of residual fixed pattern noise (RFPN) may persist in the data, negatively impacting target detection performance. In this work we examine the conditions under which RFPN can impact detection performance using data collected in the SWIR across a range of target materials. We designed and conducted a unique tower-based experiment where we carefully selected target materials that have varying degrees of separability from natural grass backgrounds. Furthermore, we designed specially-shaped targets for this experiment that introduce controlled levels of mixing be tween the target and background materials to support generation of high fidelity receiver operating characteristic (ROC) curves in our detection analysis. We perform several studies using this collected data. First, we assess the detection performance after a conventional nonuniformity correction. We then apply several scene-based nonuniformity correction (SBNUC) algorithms from the literature and assess their abilities to improve target detection performance as a function of material separability. Then, we introduced controlled RFPN and study its adverse affects on target detection performance as well as the SBNUC techniques' ability to remove it. We demonstrate how residual fixed pattern noise affects the detectability of each target class differently based upon its inherent separability from the background. A moderate inherently separable material from the background is affected the most by the inclusion of SBNUC algorithms.

Committee: Bradley Ratliff (Advisor); Jason Kaufman (Committee Member); Eric Balster (Committee Member) Subjects: Electrical Engineering

Master of Science (MS), Wright State University, 2018, Computer Science

Classification of environmental scenes and detection of events in one's environment from audio signals enables one to create better-planning agents, intelligent navigation systems, pattern recognition systems, and audio surveillance systems. This thesis will explore the use of Convolutional Neural Networks(CNN'S) with spectrograms and raw audio waveforms as inputs to Deep Neural Networks with hand engineered features extracted from large-scale feature extraction schemes to identify the acoustic scenes and events. The first part focuses on building an audio pattern recognition system capable of detecting the if there are zero, one, or two DJI phantoms in the scene within the range of a stereo microphone. The ability to distinguish the presence multiple UAV's could be used to augment information from other sensors less capable of making such determinations. The second part of the thesis focuses on building an acoustic scene detector to Task 1a in the DCASE2018 challenge(http://dcase.community/challenge2018/index). In both cases, this document will explain the pre-processing techniques, CNN and DNN architectures used, data augmentation methods including the use of Generative Adversarial Networks(GAN's), and performance results compared to existing benchmarks when available. This thesis will conclude with a discussion of how one might expand the techniques in the construction of commercial off the shelf audio scene classifier for multiple UAV detections.

Committee: John C. Gallagher Ph.D. (Advisor); Mateen M. Rizki Ph.D. (Committee Member); Thomas Wischgoll Ph.D. (Committee Member) Subjects: Computer Science

Doctor of Philosophy, The Ohio State University, 2019, Electrical and Computer Engineering

The problem of scene understanding and environment perception has been an important one in robotics research, however existing solutions applied in current Advanced Driving Assistance systems (ADAS) are not robust enough to ensure the safety of traffic participants. ADAS development begins with sensor data collection and algorithms that can interpret that data to guide the intelligent vehicle's control decisions. Much work has been done to extract information from camera based image sensors, however most solutions require hand-designed features that usually break down under different lighting and weather conditions. Urban traffic scenes, in particular, present a challenge to vision perception systems due to the dynamic interactions among participants whether they are pedestrians, bicyclists, or other vehicles. Object detection deep learning models have proved successful in classifying or identifying objects on the road, but do not allow for the probabilistic reasoning and learning that traffic situations require. Deep Generative Models that learn the data distribution of training sets are capable of generating samples from the trained model that better represent sensory data, which leads to better feature representations and eventually better perception systems. Learning such models is computationally intensive so we decide to utilize Graphics Processing chips designed for vision processing. In this thesis, we present a small image dataset collected from different types of busy intersections on a university campus along with our CUDA implementations of training a Restricted Boltzmann Machine on NVIDIA GTX1080 GPU, and its generative sampling inference on an NVIDIA Tegra X1 SoC module. We demonstrate the sampling capability of a simple unsupervised network trained on a subset of the dataset, along with proling results from experiments done on the Jetson TX1 platform. We also include a quantitative study of different GPU optimization techniques performed on (open full item for complete abstract)

Committee: Fusun Ozguner (Advisor); Keith Redmill (Advisor); Xiaorui Wang (Committee Member) Subjects: Artificial Intelligence; Computer Engineering; Computer Science; Electrical Engineering

Master of Science in Computer Engineering, University of Dayton, 2018, Engineering

Semantic segmentation has been a complex problem in the field of computer vision and is essential for image analysis tasks. Currently, most state-of-the-art algorithms rely on deep convolutional neural networks (DCNN) to perform this task. DCNNs are able to down-sample the spatial resolution of the input image into low resolution feature mappings which are then up-sampled to produce the segmented images. However, the reduction of this spatial information causes the high frequency details of the image to be lessened resulting in blurry and inaccurate object boundaries. In order to improve this limitation, I propose combining a DCNN used for semantic segmentation with semantic boundary information. This is done using a multi-task approach by incorporating a boundary detection network into the encoder decoder architecture SegNet. I explore two different multi-task learning methods of incorporating this boundary information into the SegNet architecture. These two multi-task approaches are as follows: the incorporation of the global probability of boundary algorithm and the inclusion of an edge class. In doing so, the multi-task learning network is provided more information, thus improving segmentation accuracy, specifically boundary delineation. This approach was tested on the CityScapes dataset as well as the RGB-NIR Scene dataset. Compared to using SegNet alone, I observe increased boundary segmentation accuracies using this approach. I am able to show that the addition of a boundary detection information significantly improves the semantic segmentation results of a DCNN.

Committee: Vijayan Asari Ph.D. (Committee Chair); Theus Aspiras Ph.D. (Committee Member); Eric Balster Ph.D. (Committee Member) Subjects: Artificial Intelligence; Computer Engineering; Electrical Engineering

Master of Arts (MA), Bowling Green State University, 2018, Popular Culture

The subculture of jam bands is often publicly held to multiple stereotypical expectations. Participants in the subculture are expected to fall into one of two camps, coastal elites or “dirty hippies.” Members of the Northwest Ohio jam scene often do not have the kind of economic privilege that is assumed of them based on the larger jam subculture. Not only do these perceptions create difficulties for audience members of the Northwest Ohio scene, but there are added complications for the musicians in the scene. This research explores the challenges of class and belonging faced by participants in the Northwest Ohio jam scene. More specifically, this thesis focuses on the careful social negotiations scene members and musicians are required to navigate in order to maintain insider status while dealing with the working-class realities of life in the area. In this thesis, I argue that subcultural capital is one of the most significant factors for belonging to the larger subculture, and that its necessity, which requires sufficient economic support, demands more nuanced practices by local scenesters in order to maintain. I dissect the complexities of the concept of “family” in the jam scene, including its meaning for audiences and musicians, as well as how it intersects with class and public perceptions of class in the scene. Finally, I argue that musical forms and practices hold significance in establishing genre authenticity, but I maintain that class is a determining factor in the decisions bands make about whether or not they hold completely true to genre boundaries. This thesis attempts to address the complexities of class and how it functions in small, local rock scenes, specifically in the Northwest Ohio jam scene.

Committee: Jeremy Wallach (Advisor); Esther Clinton (Committee Member) Subjects: Music