Determining Commonalities in Crowd Sourced Data
Prof. Subhasis Chaudhuri
Department of Electrical Engineering
Indian Institute of Technology Bombay
Powai, Mumbai 400 076, India
Abstract:
When several observers capture images or videos at different
times and locations, is there a common object of interest? If there is,
can we process these images to identify the common object(s) in these
scenes? This problem is known as image co-segmentation. In the first
part of the talk, we discuss how this task can be achieved. Quite
naturally, the computation goes up as the number of observations increases
albeit the size of the common object is a non-increasing function of the
number of observations. However, if these crowd sourced observations are
taken at the same geographic region and at the same time with
smart-phones, as is the norm of the day, the associated meta-data
can be used very efficiently to geo-locate the common object and to
estimate the salient view of each observer, following which
computer vision-based techniques can be used to carve out the common
object in space and time. The second part of the talk will focus on
the use of meta data.