Técnicas de limiarização para melhorar a qualidade visual de documentos históricos

Abstract

The digitization of historical documents presents itself as an effective way of enabling public access to extensive archives and equalizing the serious commitment between preservation and access. Often historical documents present severe degradation caused by effects of time or actual damage, resulting in digital images with poor legibility, and in some extreme cases, they are totally illegible. Digital image processing techniques and document analysis are employed for the solution of this problem. Most approaches to thresholding and segmentation of historical documents use global characteristics of documents, in order to eliminate background noise and other degradation problems aiming to improve legibility. However, the distribution of degradation problems is not homogeneous or linear on thedocument image. Therefore, while after the application of these algorithms some parts may present better legibility, in others the opposite effect can be observed. This work presents a new approach to the problem of image quality improvement and image legibility of historical documents. The solution uses a hybrid approach combining global and local characteristics. The processing can be divided into four stages. First, the global characteristics of the document are extracted. Secondly, the lines that present textual content areidentified and processed in the following stage. In the third stage, the thresholding of the selected lines is done combining local and global characteristics. Finally, in the last stage, global binarization of thedocument is done. Experiments done with images of historical documents from the Dops/MG archive showed that the proposed approach was efficient in improving the visual quality and legibility in 80 percent of the documents.