Social Computing for Collaborative Image Understanding

With the advance of the Internet and the increasing accessibility of computing resources, humans and computer systems are now brought together in powerful new ways. In this paper, we propose a humancentered computing framework to harness the essential characteristics of both humans and computers for achieving collaborative image understanding (i.e., training large numbers of inter-related classifiers collaboratively for automatic object and concept detection from images), where groups of volunteers may collaborate on: (a) giving their timely guidances for supporting collaborative classifier training; (b) using their personal computing resources such as PCs for training large numbers of inter-related classifiers collaboratively; and (c) assessing the correctness of learning results (classifiers and their decision boundaries) and the effectiveness of hypotheses for classifier training.