A Thresholding Approach for Text Extraction in Handwritten Historical Documents using Adaptive Morphology

The aim of preserving historical handwritten documents is to restore the degraded text containing information. But generally global threshold fails to restore the text adequately. Adaptive (local) thresholding is required for preserving the text in these documents. In recent past many standard adaptive thresholding methods have been proposed for binarization of handwritten text document images. We propose a new adaptive thresholding method using locally adaptive mathematical morphology. Formulation of an adaptive structural element is a challenging work and addressed recently by some researchers. Our method at initial step binarizes the image applying global threshold. The residual background image below threshold containing low intensity texts mixed with noise is further processed. A new approach for constructing spatially variant operator corresponding to local variances is proposed. Gaussian surface is selected as an adaptive gray-scale structuring element for mathematical morphological operations (opening and closing), whose parameters base and height depends on local variance. The proposed method successfully denoises various kinds of degraded documents enhancing textures with clear background. Experimental result on real historical handwritten document and artificial images show that our method outperforms several other existing methods both visually and using some evaluation metrics.