A highly effective approach for document page layout extraction system

In this paper, we propose a highly effective scheme for document page layout extraction system as a part of character recognition processes. There are 3 stages in the working model, namely document segmentation, document layout classification and document reading order determination. In the first stage, a hybrid document segmentation decomposes a page of the document image into a variety of blocks by using the combination of diagonal white runs and vertical edges segmentation, together with modified histogram projection. Next, the features related to geometric layout in the page are extracted by using the feature analysis, combined with the technique of rule-based approach for classifying those block types and attributes. In the third stage, a highly efficient algorithm is introduced for block order sequencing search (BOSS) as to determine the right reading sequences of blocks in the page. The model is then tested on a large number of samples of those bilingual documents with Thai and English languages, and with different geometric patterns, multiple columns, rows, fonts and sizes. The results show quite a promising one with accuracy rate of 99.47%, and the speed of 2.887 seconds per page on the average in the experiment.