KNN-LOF Algorithm Based on Skew Detection and Correction for Myanmar Handwritten Documents

Abstract

Skew detection and correction play key roles in helping character recognition processes to achieve more accurate recognition results. In this paper, we bring the idea of detection of outlier into detection skewness of Myanmar handwritten characters in printed documents. For this purpose, we present a novel detection and correction algorithm based on the k-nearest neighbor-local outlier factors algorithm. The aim is to detect local skewness that quite happens in the human's handwritten process which is in the form of fluctuation in written lines of text. Based on local skewness detection, we detect and further correct global skewness of input character image. In order to reveal the efficiency and performance of the proposed algorithm, we first prepare the datasets composed of different handwritten styles with various fonts and skew angles. We afterward perform experiments to witness how this proposed algorithm can perform well in local and global skewness detection and correction in Myanmar handwritten documents. The results show that we achieve better results in different experimental settings.