Meme Classification

The ultimate goal of this repo is to (as the title suggests) classify memes.

How it works

Currently, the the algorithm works by comparing a provided image to one of many memes provided in he comparisons/ directory. The atual comparsion is essentially just comparing the individual rgb values for each pixel after resizing the comparison image to the dimensions of the meme I would like to classify. This could loosely be defined as a euclidean distance classifier since I'm only comparing against 1 sample per class.

Hopefully this will go into further development the further my class advances into the Pattern Recognition class I'm in.

Dependencies

Both numpy and opencv are dependencies for this project. Since I have trouble installing numpy in particular to a virtualenv, I would suggest installing it into your global pip:

$ sudo apt-get install python-opencv
$ sudo pip instal numpy

If you plan to run the flask server, you will also need to install whatever is listed in requirements.txt:

$ sudo pip install -r requirements.txt

Usage

There is currently a web interface for classifying memes at http://memeclassifier.3341b.com/, but a script is also provided that can run on a bash termial.

Command line

To classify an image on your local machine:

$ python classify.py /path/to/image

To classify an image at a url, just add the --url flag:

$ python classify.py http://image.ur/path/to/image --url

Server

To start the flask server like the one that's running on the website:

$ python __init__.py

Future Work

I would like to develop this enough that I could use it as my term project for my pattern recognition class.
Instead of comparing against 1 image, collect data (from, lets say, reddit) to compare against training data via a bayesian or k-nearest-neighbors classifier.
See if it's possible to eveolve this to a point that memes could be generated from this.
If a meme is classified as a Pepe, also display it's rarity in how frequently it appears.