This was my final year project for university. It took a theory where a bulk image search and comparison could be used to give an semi-accurate attempt on where the image was taken. The code built attempts to execute on that theory.
It follows this flow
- The user uploads an image, and provides an estimated location
- A bulk image search is then made to find geotagged images that roughly match the search criteria
- Currently Flickr is the only data source, but I hope add more in the future
- It then runs a feature matching algorithm on the returned images
- It then runs a colour histogram comparison on the returned images
The improves I'd like to add in the future are
- multiple data sources to get better sources for edge cases
- better search algorithms (or use a mixture and use a knowledge-based system to determine the correct mixture of feature matchers and colour analysers)
- introduce a messaging system to facilitate the above in an async manner
- build up a record of previous successes to help in future queries