Honors Theses

Following is a list of honors theses produced by graduates of the Computer Science Bachelor of Science program. Abstracts and advisor names are included. The full thesis text is linked as a PDF file from the paper’s title, when available.

Many students presented their honors research at the Undergraduate Research Symposium until it was discontinued in 2019.

Note: Honors theses are now stored in the Carolina Digital Repository. To view more recent theses, use this link.

December 2009

Wilson Lian

Heuristic-Based OCR Post-Correction for Smart Phone Applications
Advisor: Diane Pozefsky
Optical Character Recognition (OCR) software uses computer vision and machine learning techniques to convert digital images of text into text files that may be easily modified with text editing software or used as input to a language translator or text-to-speech synthesizer. OCR software has many uses pertaining to the processing and archival of printed documents, but its application to photographic images of text, informational signs in particular, presents unique considerations and challenges. Specifically, OCR software must mitigate the effects of poor lighting, low-quality images, and unusual typefaces, all of which can contribute to character confusion and misrecognition, degrading the accuracy of the software’s output.

James Uhing

Leveraging Ontologies to Enhance Recommendation Systems
Advisor: Diane Pozefsky
Content‐based filtering techniques in recommendation engines frequently recommend pages on simple, exact tag or keyword matching. Due to the myriad of ways users might tag a page, different terms with the same or similar meaning may not be paired, while exact terms can fail to take into account semantic differences between instances of the same term. Such recommendation engines can be improved by organizing tags into an ontology. The ontology devised here consists of grouping like terms into single nodes and linking these nodes to one another through an “is a type of” relationship. In this way, we can improve relevancy scores and qualitative ratings of recommendations.

December 2008

Paul Pucciarelli

Videowner: Fingerprint Based Digital Video Comparison 
Advisor: Diane Pozefsky

May 2007

Erik Andersen

Real-time Path Planning and Simulation of Large Human Crowds
Advisors: Ming Lin and Dinesh Manocha
Human crowds are everywhere in the real world, and therefore they are important to simulate in virtual environments. We present a novel approach for real-time path planning of multiple virtual agents in complex dynamic scenes. Our algorithm is used for multi-agent planning in pursuit-evasion and crowd simulation scenarios consisting of hundreds or thousands of moving agents, each with a distinct goal.

Kris Jordan

Mixed-Initiative Access Control: Optimizing the Data Guardian’s Role
Advisor: Prasun Dewan
Robust access control capabilities are found in every modern operating-system yet end-users rarely take advantage of them. Mixed-initiative access control offers a more intuitive model for arbitrating rights between users than traditional access control.

Adam Roberts

Inferring Missing Genotypes in Large SNP Panels
Advisors: Leonard McMillan and Wei Wang
This project involved the development of an efficient technique and algorithmic implementation for inferring (filling in) missing values in Single Nucleotide Polymorphism datasets.

Joel Sutherland

Object Resource Manager and its Application to a Sea Turtle Virtual Environment
Advisor: Diane Pozefsky
This project consists of two parts. First an Object Relational Mapper was developed in PHP and second, the Mapper was used as a part of a framework to create an Experiment Management Application for the Lohmann Turtle Lab.

Adi Unnithan

Improving Search Relevancy through Human Indexing and Data Mining 
Advisor: Diane Pozefsky
Abstract: Popular search engines today index pages on a defined heuristic, such as the number of links to a page. We can improve upon this measure and display more relevant results by utilizing social networks and their repositories of information and applying data mining techniques to them, such as association rule finding and clustering. In addition, we can improve the searching experience by providing non-linearity in the user interface.