lepisma / pleiad Goto Github PK

#pleiad

Experimental word classifier based on dynamic time warping

Pleiad helps you identify words in an image. Use it when you dont need extensive character recognition and/or dont have much data, and need to identify few pre planned words only.

Pleiad doesn't need extensive training data, just a single image for each class (few more for better result) is all you need.

###Setup

• Install R (and python)
• install.packages('dtw') in R shell
• pip install -r requirements

###Usage

• Crop words from image and stretch to a fixed size (one size per classifier)

  

• Create Word objects from word images

from pleiad import pleaid
word = pleiad.Word(image, "climb")

• Train classifier from a list of Words

classifier = pleiad.PleiadClassifier(image.shape)
classifier.train(word_list)

• Predict

classifier.predict(word)

• Save for future use

classifier.save('classifierOne')

###Working

Pleiad works by treating the outer outline of each image of word as a time series and predicting by using the dynamic time warping distance between the series.

###License

MIT

Copyright (c) 2014 Abhinav Tushar