Preprocessing of Camera Captured Inscriptions and Segmentation of Handwritten Kannada text

Abstract: Deciphering ancient inscriptions on rocks, pillars and other writing material is one of the most fascinating and instructive studies and provides us with an instrument for conservation and transmission of historical traditions from generation to generation. The automatic processing of degraded historical documents is a challenge in document image analysis field which is confronted with many difficulties due to the storage condition and the complexity of their content. Enhancement of historical records which are degraded and of poor quality is not an easy task. Preprocessing is aimed at removing the background noise and improving the readability of ancient degraded documents. Thereby these enhanced document images can be transcribed easily and help in preserving the cultural heritage and historical importance.
The proposed system is designed for preprocessing ancient epigraphs and also segmentation of handwritten documents of Kannada script. Preprocessing here mainly deals with the noise removal and image enhancement of degraded ancient epigraphical images, for better human perception and also to transform the input into computer recognizable form. This is achieved through three different filtering methods for smoothing or sharpening namely Gaussian blur, Unsharp mask and Laplacian filter. These filters are used according to the varying amount and nature of image quality causing degradations. Enhancement is achieved through appropriate filter, provided with different mask sizes and parameter values which can be specified by the user. This is followed by binarization of the enhanced image to highlight the foreground information. Binarization is performed using Otsu thresholding algorithm which calculates optimal threshold value for the document image. Finally segmentation of handwritten Kannada documents is implemented using connected component method which identifies the individual characters through pixel connectivity and bounding box is enclosed around connected region to segment the character which will be displayed subsequently by the system. The output from segmentation can be used in further stages of OCR.
The proposed project works well for preprocessing of ancient documents and provides flexibility to the user in controlling the process of image enhancement to obtain desired output. Segmentation provided by the system produces better results when connectivity is present in the characters of handwritten documents.

Keywords: Epigraphical Images, Filters: Gaussian blur, Unsharp mask and Laplacian filter, Otsu method.