Bangla is a rich language with a variety of characters, which includes numerals, basic characters, compound, and modifier characters. This makes it more challenging to recognize handwritten characters in Bangla than in other languages. Pretrained convolutional neural network (CNN) architectures are effective because they can learn complex features from images, which is significant considering variations in handwriting styles. This work involves using an ensemble model consisting of pre-trained CNN architectures to classify compound bangla handwritten characters. The ensemble model comprises four pretrained convolutional neural network architectures - ResNet50, DenseNet121, Xception, and EfficientNetB0. The model is trained on MatrrivaSha dataset. To keep each image in the dataset noise-free, they are all preprocessed using a bilateral filter. The image size is reduced to 75X75 pixels, which has been used as input in several CNN architectures. Due to the limitations of RAM and GPU, the datasets are divided into subsets during model train and test.
- Python
- Numpy
- Tensorflow
- Keras
- OpenCV
- Pandas
- Seaborn
- Matplotlib
MatriVasha: Bangla Handwritten Compound Character Dataset and Recognition (https://data.mendeley.com/datasets/v39pc2g2wp/1)