Binary classification for pikachu images from other images

required python framework

sklearn, skimage, python3+, opencv, numpy

How to use

python3 train.py

'lr.pkl' will be created at 'model/lr.pkl'

test against other images:

python3 predict.py -m lr -f pikachu_dataset/extra_test/

Steps:

• keep a list all pokemon and not pokemon images (some images e.g. .gif are left out)
• extract local binary pattern features
• train with logistic regression

known issues

• gif image are left out
• some images are broken (libpng warning: iCCP: known incorrect sRGB profile)
• accuracy are very low (probably should use features better than local binary pattern and neural net instead of logistic regression)

TODO:

k-fold, against adversarial attack