[Defense] Unconventional Approaches to Color Quantization, Connected Component Analysis and Image Segmentation: Research Motivated by Making Images Accessible to Blind Students
Wednesday, May 5, 2021
9:00 am - 10:00 am
will defend her thesis
Unconventional Approaches to Color Quantization, Connected Component Analysis and Image Segmentation: Research Motivated by Making Images Accessible to Blind Students
Designing computer vision software that enables a computer to “understand” and use visual data to accomplish a task is in some ways analogous to designing software that enables a blind or vision-impaired computer user to understand imagery displayed on the computer screen. In both cases, the visual data, and the real or conceptual things it represents, has to be translated into one or more alternative forms that the unsighted consumers can take in, process, interpret and act on. In this thesis, I detail my approaches to three computer vision tasks: 1) color reduction or color quantization, 2) region finding or connected-component analysis and 3) segmentation. My approaches are, in some aspects, a significant departure from the most common methods in use today, and I will compare and contrast my approaches with those in common use. Following this, I will present my evaluation objectives, methods and results. Evaluation of segmentation has always been and continues to be a somewhat nebulous topic, and the unusual nature of my approaches only added to the challenges of designing a valid and unbiased evaluation approach. While my current implementation of the algorithms do not outperform commonly used methods in a majority of cases, I believe there is enough potential in the methods that some researchers may choose to perform follow up research. I will end my discussion with my suggestions for further research.
9:00AM - 10:00AM CT
Online via TBA
Dr. Shishir Shah, thesis advisor
Faculty, students and the general public are invited.