Plenary lecture 2: new approach for pre-processing and efficient archiving of scanned documents
Roumen K. Kountchev · International Conference on Signal Processing · 2010
Problems concerning the efficient archiving of scanned documents are some of the major contemporary challenges in this area. The standards for still image compression JPEG and JPEG 2000 are very efficient when natural images are processed, but they are not as good for texts and graphics: for relatively high compression ratios the quality of the restored images is significantly deteriorated. The efficient compression and correspondingly - archiving of documents' images requires a more flexible approach adapted to the peculiarities of the processed images. Additional problems arose when compound images, containing texts and pictures have to be processed. The best solution is each part to be processed so that to obtain maximum efficiency. A new approach for efficient archiving of scanned documents, comprising texts and pictures, is presented in this lecture. The offered approach presumes to compress the pictures and the texts in different way: the pictures - with lossy coding based on decomposition, called Inverse Difference Pyramid (IDP), and the parts, containing text (graphics) - with lossless Adaptive Run-Length (ARL) coding. The processing comprises the following main steps: -Image preprocessing, comprising background filtration (aimed at noise removal), histogram analysis and modification; -Image segmentation - recognition of texts and pictures; -Image compression - adaptive approach, which permits the pictures to be compressed with some kind of lossy IDP compression and the texts - with lossless ARL coding. The experimental results obtained for large number of example documents processed with JPEG, JPEG 2000 and the new method prove the advantages of the presented adaptive approach. The same approach is very efficient for archiving of old handwritten documents. The presented approach is based in investigations and patents developed by the lecturer and his team at the Technical University of Sofia, Bulgaria.