Applying Image Enhancement Techniques for Improving the Quality of Historic Amharic Manuscripts

Abstract

A large number of manuscripts are found in various institutions, archives and other museums and it is common for archive libraries to provide public access to historical and ancient manuscript digital image collections to protect original hard copies from damage and also to improve access. The Ethiopian National Archives and Library Agency (ENALA) is in the process of digitizing its historical manuscript collections and there is a need to ensure the quality of the digital images of the manuscripts. This study is initiated with the objective of testing various image noise filtering algorithms for their effect on enhancing digital images of manuscripts under different conditions. Six historical manuscripts from the early years of the 20th Century and two from each of manuscripts in good, medium and poor conditions were purposely selected for this study. Images of each manuscript were taken using digital cameras with 8 and 16 megapixel resolution. The images were filtered using mean, median, Gaussian, morphological dilation and morphological erosion algorithms with convolution kernel size of 3 by 3 and 5 by 5. The PSNR, MSE and MAE are used as parameters of evaluation of the effectiveness of the various filtering algorithms. The result has shown that similar trend of the three parameters for all images and filtering algorithms. In addition correlation values of higher than 0.87 were found for the parameters between themselves. For the first set of image PSNR values of 21.64, 19.14, 16.78, 12.89 and 12.60 are obtained for the median, mean, Gaussian, morphological dilation and Morphological erosion algorithms. Similar trend was observed for most of the other images. No consistent effect of image resolution was observed but images of 8 megapixel have shown better quality after filtering images of manuscripts in good and poor conditions, whereas images from manuscripts of medium condition with 16 megapixel has produced better images upon filtering. Kernel size at 3 by 3 was found to be better than 5 by 5 for enhancing image quality. From the experimented filtering algorithms, median filtering algorithm with 3 by 3 kernel size is proved effective to improve the quality of images of historical manuscripts. Digital images 2 taken at a minimum resolution of 8 megapixels appear to be adequate for images of historical manuscripts. Evaluation of advanced image filtering techniques is recommended for improving the quality of highly degraded manuscript images.