Multi-level Skew Correction Approach for Hand Written Kannada Documents

Abstract

During capturing documents using camera or camera phone and scanning the documents, document skew is unavoidable. Skew in line and paragraph of handwritten document is varying for different people. Document analysis and character recognition efficiency is mainly depending on document pre-processing, document skew correction. Document skew detection and correction is one of the difficult step before document analysis. In this paper we proposing multi-level Run-Length-Smoothed Image (RLSA) skew detection and correction for kannada hand written document image. We have major two sections, first section is pre-processing techniques like Haar wavelet decomposition, maximum gradient, 4, 8 connective Laplacian methods to extract text region without losing data. Second section is multi-level skew detection, horizontal and vertical projection profile are used to detect the page boarder and further RLSA skew detection is applied to find skew angle and rotate document by desired angle, this removes the document skew occurred while capturing the image. Later, the skewed document will be further processed to remove line and paragraph skew angle by applying RLSA skew detection technique for each segmented line and rotate each line by desired angle. Performance of proposed system is performed for kannada hand written documents; the experimental results are discussed; the proposed system is encouraging.