Offline Handwritten Kannada Sentence Recognition Using Knn With Convolutional Neural Network

Abstract

Multiple algorithms have been proposed for character segmentation, recognition of multilingual Indian document images of Devanagari scripts. This document is suffering from its outer layout organizations, different handwriting styles, local skews, bad print quality, and has overlapped texts. There are three levels of segmentation namely line, word, character segmentation in the suggested segmentation algorithm. Using the structural character property segmentation paths are obtained, and graph distance theory is using joined and overlapped characters are divided. The highly accurate KNN classifier is used to certify the segmentation consequences. In the Proposed character recognition algorithm, the CNN is used to train and test the handwritten dataset with the help of the KNN classifier increases the accuracy for testing the model and an accuracy of 92.49 % is achieved.   Several experiments have been made on various databases including printed just as manual texts. Benchmarking results tell that proposed methods for recognition of Kannada handwritten sentence recognition have better performances in terms of segmentation of lines, words, and characters related to different methodologies, where the most elevated segmentation and recognition rates are achieved.