基于犹豫模糊集的古籍汉字图像切分方法

针对古籍文献版面图像切分中存在的过切分和粘连等问题,提出基于犹豫模糊集的古籍汉字图像切分方法。首先,对古籍汉字版面图像进行连通区域搜索,获取版面中的笔画部件,实现古籍汉字的初切分;然后,对过切分汉字进行特征分析,提取过切分区域的特征,利用犹豫模糊集在处理多属性决策问题方面的优势,通过建立犹豫模糊集来判断过切分区域归属同一古籍汉字的隶属度,并据此进行过切分区域的合并;最后,对存在粘连和重叠的汉字采用分段像素跳跃数突变分析方法进行分割。在28 886个古籍汉字上的实验结果显示,可以达到92. 3%的切分准确率和85. 7%的过切分合并准确率。

Segmentation method of Ancient Chinese Character Images Based on Hesitant Fuzzy Set

In order to solve the problems of over-segmentation and cohesion in the layout segmentation of ancient books and documents, a segmentation method of Chinese characters in ancient books based on hesitant fuzzy sets was established. Firstly, to realize the originally segmenting of Chinese characters in ancient books, the connected area search was carried out to obtain stokes in layout; Secondly, the feature analysis of over-segmentation Chinese characters was performed to extract the feature information of over-segmentation regions. Based on the advantages of the hesitant fuzzy sets in dealing with multi-attribute decision-making problems, a hesitant fuzzy sets was established to judge the membership degree of over-segmentation regions belonging to the same ancient Chinese characters, then merged the over-segmentation regions; Finally, the Chinese characters with adhesions and overlaps are segmented by using segmented pixel jump number mutation analysis method. In conclusion, the experimental results on 28 886 ancient Chinese characters show that the accuracy of segmentation can reach 92.3% and the accuracy of the over-segmentation is 85.7%.