Candidate Eyegaze and Manual Input Methods for an Improved User Experience in Interactive Image Segmentation
Evan Strasnick, Szymon Rusinkiewicz
Princeton University, 2014.
Best Poster Award
While numerous efficient algorithms exist for both interactive and non-interactive image segmentation, the lack of a useable interface has hampered their use in consumer applications. This paper examines a number of candidate input schemes using both mouse and eyegaze control on a seed-based image segmentation task, and measures quantitative and qualitative considerations in providing an accurate, efficient, and usable approach to segmentation. Our results indicate that the primary factors determining accuracy and perceived usability do not include the specific scheme by which users place seeds, but rather the degree of precision and control afforded to the user by a particular scheme or device. Further, users tend to approach segmentation tasks with a few recognizable strategies, even if suboptimal for the given algorithm/input scheme, and therefore greater attention should be given towards designing segmentation tools with users’ natural tendencies in mind.