An application for classifying new indian currency notes using random forest classifier. It is implemented with the help of opencv and scikit-learn library. The dataset used for training consists of 1050 images of new currency notes(INR). It includes denominations of 10, 20, 50, 100, 200, 500 and 2000 rupees notes(i.e. 7 classes with 150 each).
Here we use hu moments(shape), haralick(texture) and colour histogram(colour) as global features and bag of visual words(BOVW) with SIFT as a local feature descriptor. It is then trained using a random forest classifier using scikit-learn library.
Finally during inference, we extract the rectangular ROI's from the preprocessed image using opencv(cropped rotated contours) and predicts the class labels using the trained random forest model.
- Python3, Scikit-learn
- Opencv+contrib, Mahotas
- Pickle, Joblib
python bovw.py # Feature extraction
python hyper_train.py # Hyperparameter tuning
python train.py # Model training
python predict.py # Model inference
python currency.py # Currency classificationBest Parameters:-
bootstrap=True, criterion='gini',
min_samples_leaf=1, min_samples_split=2,
min_weight_fraction_leaf=0.0, n_estimators=200,
n_jobs=-1,verbose=1
Cross Validation Accuracy: 0.98 (+/- 0.01)
Confusion Matrix:-
| 10 | 20 | 50 | 100 | 200 | 500 | 2000 | |
|---|---|---|---|---|---|---|---|
| 10 | 28 | 0 | 1 | 0 | 2 | 0 | 0 |
| 20 | 0 | 24 | 0 | 0 | 0 | 1 | 0 |
| 50 | 0 | 0 | 27 | 0 | 0 | 0 | 0 |
| 100 | 0 | 0 | 0 | 32 | 0 | 0 | 0 |
| 200 | 2 | 0 | 0 | 0 | 31 | 0 | 0 |
| 500 | 0 | 1 | 0 | 0 | 0 | 26 | 0 |
| 2000 | 0 | 0 | 0 | 0 | 0 | 0 | 35 |
Version 1.0
Anil Sathyan
- "https://kushalvyas.github.io/BOV.html"
- "https://gogul.dev/software/image-classification-python"
- "https://github.com/briansrls/SIFTBOW"
- "https://www.pyimagesearch.com/2017/01/02/rotate-images-correctly-with-opencv-and-python/"
- "https://towardsdatascience.com/optimizing-hyperparameters-in-random-forest-classification-ec7741f9d3f6"
- "https://stackoverflow.com/questions/11627362/how-to-straighten-a-rotated-rectangle-area-of-an-image-using-opencv-in-python/48553593#48553593"
- "https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_contours/py_contour_features/py_contour_features.html"