Projects
Recovery of Damaged Text in Cultural Heritage Imaging
My senior Imaging Science Project was implementing computer vision methods to recover damaged text in cultural heritage documents. I worked with my advisors Dave Messinger and Roger Easton.
Abstract: Reagent-treated documents are plentiful, due to the use of chemicals on palimpsest documents to improve the legibility (albeit temporarily) in the 19th century. The text damaged to the point of unreadability on these documents is of great interest to scholars. We propose a method for extracting text pixels from multispectral images of reagent treated documents. Using nonlinear manifold modeling to reduce the dimensions of the image data, the pixels contributing to text can easily be detected. In four steps: Converting the Image to a Graphical Network, Graph Schrodinger, Eigenvalue Problem and Detection. This method shows promising results in simple cases. Further work is required to better understand how well the method works on sections of documents heavily treated with reagent.
Imaging Palimpsests Presentation
In May of 2021, a couple of my co-workers and I presented at the International Congress on Medieval Studies (ICMS). Our presentation was on the imaging system we built for imaging removed text on historical documents and the set of Ege leaves that we discovered were palimpsests in RIT's Cary Graphic Arts Collection. We worked on this project for over a year by the time we presented at the ICMS, so we had given plenty of presentations on our work in the past. The more presentations we gave the better we got at creating and presenting them. In general, I try to use more visuals than words on my slides. Instead of the audience reading the slides while I speak I want the slides to have visuals that enhance the understanding of what I'm explaining. My presenting skills have also improved. I'm more comfortable and less nervous presenting than I was a couple of years ago. I make sure I'm looking at the audience, not reading off the slides. I'm prepared for presentations so I don't get stuck or lost while presenting.
Imaging Palimpsests Interviews
My coworkers and I gave many presentations with slides, but we also were interviewed a few times. This format is similar to presenting but it's harder to prepare for since you may not know what you will have to talk about specifically. What made interviews easy for me was that I loved the work I was doing. I felt confident that I knew what I was talking about and wanted to share my work and discoveries with others. On the right are two interviews that we did, one on Even Dawsons NPR show Connections and one for RIT News.
Open CV Tracking Algorithms and Sport Players
Analytics in all kinds of sports are used to improve the performance of players. Collecting and analyzing player data by hand can be time-consuming and costly. This paper analyzes the effectiveness of built-in OpenCV tracking algorithms on player tracking. Based on the high error rate it is not recommended that any of the 8 OpenCV tracking algorithms be used for player tracking, but MOSSE and Boosting performed the best out of all the methods tested.
​
This paper was part of an independent study completed in spring 2022. Code referenced in the paper, and for the independent study as a whole, was written in Python.
WIC Hacks - Overhaul Ordering
RIT's Women in Computing club (WIC) puts on a hackathon every year called WIC Hacks. In February 2021 I participated in WIC Hacks on a team with one other person, who was a computer science major. Since I had imaging science and computer vision skills, and my teammate had data analysis skills we decide to combine our skills to create something unique. Amazing things can be accomplished when you combine your skills with someone else's. We created a program that would gather order information by imaging an order slip and then would add that information to display for the cooks. This display would tell the cooks how many of each item needed to be made, so order information could be concise. Programs like ours show that with some collaboration we can solve everyday problems. Our final program was called Overhaul Ordering or O² and we won M&T Bank's Best Digital Business Accelerator prize. On the right, you can go to a page that explains more about the project and has links to the code as well as the presentation we gave at the hackathon.
Color Transfer
This video goes through the process of transferring color between images. This method is based on the paper Color Transfer between Images by Erik Reinhard et al.
​
This video was created for the course Image Processing and Computer Vision II. The code for color transfer was written in Python.
Seam Carving
Images have to fit on various screen sizes, from wide TV screens to skinny and tall phone screens. Cropping images to a desired size usually results in the removal of wanted information near the edges of the image. An image’s aspect ratio can also be changed to a desired ratio, but this can cause objects in the image to look squished or stretched. In order to preserve the ”important” aspects of an image, the content needs to be taken into account when resizing. The proposed method uses seams to systematically add or remove pixels in connected lines from an image without removing important image content. Seams are paths that go from one side of the image to the opposite side. The path is determined by the energy of the pixels. With the seam method of resizing images, areas with edges are preserved while areas with great similarity can be added to or removed from. This results in a resized image that preserves the look of features with lots of edges.
This paper was created to go along with code written for a course called Image Processing and Computer Vision II. The code for this project was written in Python.
CFA Interpolation
Color filter arrays create the need for the interpolation of missing values in the three color channels. Simple and common methods like bilinear and Laroche and Prescott interpolation tend to fail at the edges in images. The method this paper proposes improves the reconstruction of edges compared to other methods. This is achieved by using a narrow edge detector and a strong focus on green channel reconstruction.
This paper was created to go along with code written for a course called Image Processing and Computer Vision II. The code for this project was written in C++.