This is an experiment to use OpenCV to detect duplicate images. It's a purely research project not intended to be used in a production enviroment with big data sets.

This is a standard CMake project and build procedure is quite straightforward. Assuming that the source files are located in the /path/to/imgdupl-opencv/ directory one needs to perform following steps:

$ mkdir /path/to/imgdupl-opencv-buildroot/
$ cd /path/to/imgdupl-opencv-buildroot/
$ cmake /path/to/imgdupl-opencv/ -DCMAKE_BUILD_TYPE=Release
$ make

First you need to create a database with extracted keypoints and features of an image set. To do this you use build-imgdupl-db utility. For a while invocation is very simple and basically looks like:

$ ./build-imgdupl-db directory database

Where directory contains images and database is a file in filesystem. Utility recursively walks through directory tree and fills database with corresponding data. We use SQLite engine for data management.

Example:

$ ./build-imgdupl-db /mnt/disk/Photos/Party/ /tmp/images.db

Example:

$ ./imgdupl --db-file /tmp/images.db --config imgdupl.cfg

$ curl --data-binary "@file.jpg" -H "Content-Type: application/octet-stream" -X POST http://127.0.0.1:9090/check

Here file.jpg is a name of a file on the disk, i.e. /full/path/to/image_001.jpg.