摘要
甲骨拓片图像往往遭受点状噪声、人工纹理和固有纹理等的严重影响,导致典型的图像分割方法尚无法得到令人满意的文字分割结果.为此,提出一种连通区域拓扑结构约束的甲骨拓片图像分割方法.该方法采用逐步求精的层次分割策略,首先,利用超像素分割和最大类间方差法进行初始分割;其次,采用最小凸包算法分离断裂的甲骨文字与粘连的拓片背景,得到粗分割结果;最后,利用形态学运算、欧拉数、方向投影变换对连通区域的拓扑结构和形状特征进行分析,得到精细分割结果.实验结果表明,在多种典型的图像分割方法中,该方法在正确分割率、敏感度、特异度3个指标上较最优方法分别提升了2.03%、0.23%、1.54%.并且,建立连通区域的形态学先验是克服背景粘连、大面积片状斑纹和盾纹的影响及改善甲骨拓片图像分割效率的有效手段.
The oracle bone rubbing images are often severely affected by point noises,artificial textures,as well as inherent textures.It results in that state-of-art image segmentation methods cannot achieve desirable segmentation results yet.Hence,this study proposed an image segmentation method for oracle bone rubbings with topological structure constraint of connected regions.It adopted a hierarchical segmentation strategy in a coarse-to-fine manner.The super-pixel segmentation and the Ostu method were first used to complete coarse segmentation.Then,the minimum convex hull algorithm was employed to separate the broken oracle bone characters and adhesive rubbings background.Finally,the topological structures and shape features of the connected region were analyzed by exploiting the morphological operations,Euler numbers,and directional projection transform,obtaining the refined segmentation result.Experimental results show that the proposed method has separately improved the segmentation accuracy,sensitivity,and specificity by 2.03%,0.23%,and 1.54%compared to the optimal method,among various typical image segmentation methods.Moreover,it is an effective means to establish morphological priors of connected regions to overcome background adhesions,large area patchy and shield patterns,and to improve the segmentation efficiency of oracle bone rubbing images.
作者
宋传鸣
周雨晴
张晋豪
洪飏
SONG Chuanming;ZHOU Yuqing;ZHANG Jinhao;HONG Yang(School of Computer Science and Artificial Intelligence,Liaoning Normal University,Dalian 116029,China;School of Chinese Language and Literature,Liaoning Normal University,Dalian 116029,China;Provincial Key Laboratory for Computer Information Processing Technology,Soochow University,Suzhou 215006,China)
出处
《闽南师范大学学报（自然科学版）》
2023年第4期35-50,共16页
Journal of Minnan Normal University：Natural Science
基金
辽宁省社会科学规划基金资助项目(L19BYY005)。
关键词
甲骨拓片
图像分割
数字图像处理
连通区域
拓扑结构
形态学
oracle bone rubbings
image segmentation
digital image processing
connected region
topological structure
morphology
