A novel method for segmenting and straightening of text lines in handwritten Telugu documents based on smearing and regression approach

Abstract

In handwritten document images, segmenting text lines is a very challenging task due to various reasons like variability in intra baseline skew and inter line distance between text lines. So far, no work is reported in the literature for the straightening of handwritten Telugu languages. Telugu is one of the most popular languages of India that is spoken by more than 80 million people especially in South India. Telugu characters are mostly compound characters and that is way the straightening task of Telugu document is more challenging tasks than European languages. This paper introduces a novel approach for segmenting and straightening text lines of handwritten Telugu documents based on smearing and regression approach (SRA). This method initially performs preprocessing and estimates parameters by dividing into connected components of Telugu script. A horizontal and vertical run length-smearing algorithm is used in this paper to shape text lines. To identify text lines more precisely cubic polynomial regression is used between vertical midpoints of two blocks of compound handwritten Telugu characters. A simple logic is derived on this to achieve final process. We tested the proposed algorithm with three different kind of 1000 handwritten documents. The performance of proposed method is evaluated by using matchScore, detection rate, recognition accuracy and F-measure. The experimental results indicates the efficiency of the proposed method over the existing methods.