Intelligent document processing system for conference article

摘要

The conventional document processing systems include document analysis (DA), document classification, and document understanding. These systems are step by step. If the results in the previous step are improper, the current step will produce improper results. Furthermore, the binarization methods in DA to threshold an A4-sized color image are inefficient because they scan the entire image at least once. The block segmentation methods in DA to segment an A4-sized binary image are inefficient since they scan the entire image at least twice. The layout analysis methods in DA are also inefficient. They use global and local analysis and scan the entire image at least once. In this article, an intelligent, efficient, and effective document processing system is proposed to solve the abovementioned problems. The proposed method includes document binarization and mixed-based layout analysis. The binarization method only scans the border image. The mixed-based layout analysis mixed uses block segmentation and classification. The block segmentation only scans the background image. The block classification uses background gap and writing format to classify blocks. Experimental results show that the performance of the proposed method is better than FineReader 11.0 in visual measurement.