One of the challenges in image processing in a law enforcement context is that of understanding spatio-temporal context.
This problem concerns not only inferring the topic of the material, but addressing ‘where and when was this material produced’, ‘is it part of a series’, and/or ‘have we seen material from a similar place or time before’? This inference needs to be done by extraction and recognition of relevant features from the content of the image/video itself, rather than metadata such as EXIF which may be missing, obscured, or incorrect. How do we move beyond detecting features to inferring aspects of their meaning?
This project is working toward identifying novel methods for inferring where and when photographs and videos were recorded from features of the material itself.