paapu88 / Plate

Plate

from an image by haar cascade get plate, then from plate get regions (by MSER) , from regions get letters/digits by SVM/logistic regression or by neural network

Usage:

 python3 image2characters.py "filename"

NOTE:

the path one above current directory (Plate) must be in $PYTHONPATH
for neural network you need weight.npz

Background:

Haar cascade description:

https://archive.ics.uci.edu/ml/datasets/Letter+Recognition raw training data in mka@mka-HP:~/PycharmProjects/Rekkari/Training

The SVM/logistic regression files are trained in

└── TrainSVM ├── Digits │   └── SvmDir ├── Letters │   └── SvmDir

copied as follows Kauppi:/PycharmProjects/Image2Characters> cp TrainSVM/Letters/SvmDir/logistic.pkl letters_logistic.pkl Kauppi:/PycharmProjects/Image2Characters> cp TrainSVM/Letters/SvmDir/allSVM.txt.dict letters_logreg.dict Kauppi:/PycharmProjects/Image2Characters> cp TrainSVM/Digits//SvmDir/logistic.pkl digits_logistic.pkl Kauppi:/PycharmProjects/Image2Characters> cp TrainSVM/Digits//SvmDir/allSVM.txt.dict digits_logreg.dict

Sphinx:

sphinx-quickstart

autodoc: automatically insert docstrings from modules (y/n) [n]: y edit index.rst make html