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

Evaluating anomaly factors in images is a pivotal element in advancing the robustness of image processing techniques, particularly under adverse and dynamic conditions. This thesis presents a comprehensive investigation into anomaly factors, focusing on two major evaluations: anomaly addition and anomaly removal. In the first evaluation, anomaly addition, we assess the resilience of computer vision frameworks in real-world scenarios. Specifically, this involves studying the performance of object detection algorithms in adverse weather conditions, such as fog, rain, snow, and sun flare, which pose significant challenges to autonomous vehicle technologies. Our methodology includes calculating Intersection over Union (IoU) to measure bounding box overlap between model predictions and ground truth labels, allowing for an accurate assessment of true positives (TP), false positives (FP), and false negatives (FN) across multiple classes. We use performance metrics such as class accuracy, precision, recall, F1 score, and average accuracy to provide a comprehensive view of model robustness. Through ablation studies and dual-modality architecture analysis, the impact of these anomalies on traffic monitoring, vehicle tracking, and object detection is thoroughly examined. The findings underscore the limitations of algorithms trained under clear weather conditions and emphasize the need for more adaptive systems to ensure safety and efficiency in intelligent transportation technologies. The second evaluation, anomaly removal, explores the effectiveness of image inpainting techniques in removing undesired elements, such as photobombing, from images. A benchmarking study was conducted to compare state-of-the-art inpainting methods on a dataset of over 300 images. Using performance metrics like PSNR, SSIM, and FID, the results reveal both the strengths and limitations of current techniques in restoring images with varying levels of complexity. Our evaluation provides a valuab (open full item for complete abstract)

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

Master of Science (M.S.), University of Dayton, 2020, Electrical Engineering

Convolutional neural networks (CNN) are revolutionizing and improving today's technological landscape at a remarkable rate. Yet even in their success, creating optimal trained networks depends on expensive empirical processing to generate the best results. They require powerful processors, expansive datasets, days of training time, and hundreds of training instances across a range of hyperparameters to identify optimal results. These requirements can be difficult to access for the typical CNN technologist and ultimately wasteful of resources, since only the most optimal model will be utilized. To overcome these challenges and create a foundation for the next generation of CNN technologist, a three-stage solution is proposed: (1) To cultivate a new dataset containing millions of domain-specific (aerial) annotated images; (2) to design a flexible experiment generator framework which is easy to use, can operate on the fastest supercomputers in the world, and can simultaneously train hundreds of unique CNN networks; and (3) to establish benchmarks of accuracies and optimal training hyperparameters. An aerial imagery database is presented which contains 260 new cultivated datasets, features tens of millions of annotated image chips, and provides several distinct vehicular classes. Accompanying the database, a CNN-training framework is presented which can generate hundreds of CNN experiments with extensively customizable input parameters. It operates across 11 cutting-edge CNN architectures, any Keras-formatted database, and is supported on 3 unique Linux operating systems - including two supercomputers ranked in the top 70 worldwide. Training can be easily performed by simply inputting desirable parameter ranges in a pre-formatted spreadsheet. The framework creates unique training experiments for every combination of dataset, hyperparameter, data augmentation, and super computer requested. The resulting hundreds of trained networks provides the performance to perform (open full item for complete abstract)

Committee: Eric Balster (Committee Chair); Patrick Hytla (Committee Member); Vijayan Asari (Committee Member) Subjects: Artificial Intelligence; Computer Engineering; Computer Science; Electrical Engineering; Engineering

Master of Computer Science (M.C.S.), University of Dayton, 2020, Computer Science

Text-to-image is a process of generating an image from the input text. It has a variety of applications in art generation, computer-aided design, and photo-editing. In this thesis, we propose a new framework that leverages mask anchor points to incorporate two major steps in the image synthesis. In the first step, the mask image is generated from the input text and the mask dataset. In the second step, the mask image is fed into the state-of-the-art mask-to-image generator. Note that the mask image captures the semantic information and the location relationship via the anchor points. We develop a user-friendly interface that helps parse the input text into the meaningful semantic objects. However, to synthesize an appealing image from the text, image aesthetics criteria should be considered. Therefore, we further improve our proposed framework by incorporating the aesthetic assessment from photography composition rules. To this end, we randomize a set of mask maps from the input text via the anchor point-based mask map generator, and then we compute and rank the image aesthetics score for all generated mask maps following two composition rules, namely, the rule of thirds along with the rule of formal balance. In the next stage, we feed the subset of the mask maps, which are the highest, lowest, and the average aesthetic scores, into the state-of-the-art mask-to-image generator via image generator. The photorealistic images are further re-ranked to obtain the synthesized image with the highest aesthetic score. Thus, to overcome the state-of-the-arts generated images' problems such as the un-naturality, the ambiguity, and the distortion, we propose a new framework. Our framework maintains the clarity of the entities' shape, the details of the entity edges, and the proper layout no matter how complex the input text is and how many entities and spatial relations in the text. Our contribution is converting the input text to an appropriate constructed mask map or to a set (open full item for complete abstract)

Committee: Tam Nguyen (Advisor) Subjects: Computer Science