基于图理论学习模型的胃窥镜图像自动标注

Automatic Endoscopic Image Annotation Based on Graph Learning Model

为了解决目前医学图像检索领域不能有效缓解"语义鸿沟"的问题,提出基于图理论学习模型的图像自动标注方法.首先讨论了医学图像的标注问题,总结了现有关医学图像标注的研究工作.以胃窥镜图像为具体研究对象,针对图学习模型中的图像-标注词间的关系提取以及图像相似度计算进行了详细分析,并有效地融合进医生的诊断信息作为图像的高级语义特征,更有效地计算出图像间相似度.最后,在Toy data数据集和临床胃窥镜图像集上进行了一系列的实验,结果表明本文方法优越于传统图像标注方法.

To solve the ＂semantic gap＂ problem in medical image retrieval, the paper proposed the automatic image annotation based on graph learning. It discussed the process of medical image annotation, and summarized related researchworks. Choosing endoscopic images as the object, the thesis analyzed the ectraction of the relationships between images and annotation words as well as the image similarity computation, compromised doctors＇ diagnostic information as the high-level semantic features of the images, which effectively calculated the image similarity. A series of experiments were conducted on Toy data and endoscopic images, the results show the method in this paoer is better than the traditional image annotation methods.