Abstract: Aiming at the feature that transverse documents and vertical documents blend mostly in Chinese document layout, a menthod based on minimal spanning tree clustering is presented. Apply run_length smoothing algorithm on the document in horizontal direction, and vertical direction. Then, a pre_classification step is applied to the connected components generated after classifying run_length smoothing to body text into horizontally aligned and vertically aligned. Minimal spanning tree clustering algorithm is applied to the body text that are generated after pre_classification. Via experiment, the accurate rate reaches 97%. As is shown from the experiment, the method has a good effect on segmentation of Chinese documents.

Key words: layout segmentation, run_length smoothing, minimal spanning tree clustering

摘要： 针对中文版面多横竖混排的特点，提出一种基于最小生成树聚类的版面分割方法。对原图像进行水平和垂直游程平滑，并对平滑后所得的连通域进行预分类处理，将文本进行横排、竖排分类。对预分类后的各类文本采用最小生成树聚类算法进行聚类处理。经实验，准确率达97%。实验表明，该方法对中文文档有良好的分割效果。

关键词: 版面分割, 游程平滑, 最小生成树聚类

CLC Number: 

• TP391.1