Principal Component Analysis Research at Hiram College Inspires Computer Vision Project at The University of Hartford
Every computer science student at Hiram College is required to perform two research projects, called IRCs, prior to graduation. Of the two projects I completed I am proudest of my Computer Vision project because it not only was one of the coolest things I’ve ever studied, but it also required an extensive amount of hard work to complete. The satisfaction is still a source of motivation for me to this day.
The project’s in-depth details can be found at http://www.geigel.com/signlanguage/, however, briefly, I was able to employ computational methods, initially researched in the 1980s, to recognize American Sign Language letters within computer images. (Note: The initial research performed in the 1980s focused on recognizing faces in images which have popularly become known as Eigenfaces). The process is called Principal Component Analysis (PCA) and is widely considered to be a breakthrough discovery in the advancement of Computer Vision. Much of PCA’s attractiveness comes from it’s ability to train similar, yet slightly different, images of a given class. In my research, for instance, the classes I dealt with were signed letters corresponding to different hand orientations. When it boils down to it, PCA is able to mathematically define (using Eigenvectors and Eigenvalues) the most important features of a given class (for example a fist for letter A versus the open cirlce for the letter O), and then when a new unknown image is presented to the system a calculation can be performed and a mathematically reinforced estimate can be made as to what class the unknown image belongs.
Ellen Walker, my Computer Vision professor at Hiram College, oversaw my research and offered help when I ran into issues. Ultimately, I was able to develop/program/execute a fully functional PCA system in C++ for recognizing American Sign Language letters. In fact, once the PCA system was trained, I was able to test it on 10 sample images with 100% recognition accuracy. The project was a huge success! Both in it’s results and in the lessons, knowledge, and gratification it left me. Though Ellen has never explicitly said this to me, I think that she was also impressed and surprised with how well the results turned out.
This all took place in 2005 — the year I graduated from Hiram. As the years went on I would occasionally correspond with Ellen and then on 11/29/2006 I received an email from her asking if my project could serve as a possible AI/Vision project that she would be submit to the University of Hartford’s “Machine Learning Experiences in AI” shared curriculum. Of course I was excited and happy that my project would be the inspiration for such a project.
A few years later I received another email from Ellen on 2/23/2009 saying that project was now appearing on Hartford’s website at http://uhaweb.hartford.edu/compsci/ccli/samplep.htm. The project’s detail page can be found at this URL: http://uhaweb.hartford.edu/compsci/ccli/rasl.htm. There’s also a PDF (http://cs.hiram.edu/~walkerel/RASLUPCA.pdf) that has in-depth details along with a credit at the very bottom stating that the project was inspired by my project! In classic programmer speak — w00t!