Robust recognition technique for handwritten Kannada character recognition using capsule networks

<p>Automated reading of handwritten Kannada documents is highly challenging due to the presence vowels, consonants and its modifiers. The variable nature handwriting styles aggravates complexity machine based vowels consonants. In this paper, our investigation inclined towards design a deep convolution network with capsule routing layers efficiently recognize characters. Capsule architecture built an input layer, two layers, primary capsule, followed by tri-level dense layer output layer. For experimentation, datasets are collected from more than 100 users for creation training data samples about 7769 comprising 49 classes. Test all classes again separately 3 5 creating total 245 novel patterns. It inferred performance evaluation; loss 0.66% obtained in classification process 43 precision 100% achieved accuracy 99%. An average 95% remaining 6 89%.</p>