Abstract

With the advent of image recognition using deep learning, various applications of it are put forth. Optical Character Recognition as well as Handwritten Character Recognition is at the forefront of it. For this reason, optimized deep learning models are required to increase the accuracy of recognition. One such model is Capsule network. Instead of traditional CNN s which causes invariance, capsule networks provide equivariance. In terms of OCR, spatial arrangement of the feature is more important than features themselves. But capsule networks are computationally heavy. Hence, by applying pruning to a CapsNet model, we can reduce the computational costs and time complexity.