A Novel Approach to Recognize and Validate Kannada Numerical using Capsule Networks
Abstract
A very important field of computer science is recognition of handwritten text. With the advancement of machine learning and deep learning techniques there has been a lot of research in this field. The existing methods can still be improved. This paper proposes a novel system for recognition of Kannada numbers using capsule nets which is an advancement over Convolutional neural Network (CNN). A Kannada MNISTdataset is used to evaluate the system which consists of 60000 grey scale images. The use of capsule nets overcome drawbacks of CNN by accounting for the pose of features in an image. Capsule nets establish spatial relationships between features and use the dynamic routing by agreement algorithm to pass data between layers. Thus excellent results have been obtained by this system with respect to all baseline methods.