Quantum-Classical Framework for Tamil Handwritten Character Classification using SQCNN and Bayesian-Optimized VQC

Handwritten character recognition for Tamil characters is challenging because of the characters displaying high level of similarity between them and also due to the variations in different handwriting styles. Quantum CNN models have been explored recently in character recognition tasks especially when the amount of training data is limited. Quantum models have the ability to encode and process the information in higher dimensional spaces using quantum properties like superposition and entanglement. In this paper, a hybrid quantum-classical framework has been proposed which is used for classification of ten handwritten Tamil characters. This hybrid framework consists of a Scalable Quantum Convolutional Neural Network which does patch wise local feature extraction and a Bayesian-optimized Variational Quantum Circuit which is used as a global feature extractor and then final classification is done using a classical fully connected layer. This hybrid model achieves better classification accuracy for Tamil handwritten characters while reducing the total number of trainable parameters required to train the model.