Automated face-based identification of Website content (Dr. Terry Boult)
The second biometrics-based project will be focused on building tools for fusing image-based and text-based analysis to help identify individuals in photographs that appear in photographs on web pages. The research questions for the student to explore are building probabilistic face-based algorithms models that exploit constraints obtained from the text based processing, as well as text-processing techniques that can exploit potential hypothesized identities or properties (e.g. gender) obtained from the vision algorithms. This effort is synergistic with the above effort, as improved automation of this process can improve the web-based crawler for building a large database, and the larger database can improve the recognition core of the automated processing. The key issue here is face-recognition in context where the text will provide potential identification labels for subjects and may also provide keys for co-factor identification, e.g. recognition of pronoun can help with gender.