Written as part of a class project on near-duplicate detection for videos, this project is a Matlab script that uses Caffe neural networks to generate descriptive features for video frames. These descriptive features provide a unique signature for videos that can be queried to find similar or near-duplicate videos, even after various distortions, including resizing, rotating and cropping.
A summary of the technique and results can be found in the report.
L. Nathan Perkins