文献类型：专利

英文题名：Visual semantic graph based image area labeling method, involves analyzing global similarity between image areas, and performing un-marked image updating process between image region label and image areas for labeling un-marked image

作者：ZHANG J;YAO D;MU Y;WANG Z;ZHAO X;CHEN M

机构：[1]UNIV EAST CHINA SCI & TECHNOLOGY

申请号：CN107967494-A

申请日：2017-12-20

公开日：2018-04-27

语种：英文

收录：DERWENT

摘要：NOVELTY - The method involves analyzing global similarity between image areas. An image region label semantic association process is performed. Relationship between the image areas is analyzed to form a visual semantic graph, where a visual semantic graph comprises a global similarity image graph and an image region similarity graph. An un-marked image is matched into the visual semantic graph by using semantic graph across hierarchical random walk algorithm. An un-marked image updating process is performed between the image region label and the image areas for labeling the un-marked image. USE - Visual semantic graph based image area labeling method. ADVANTAGE - The method enables realizing semi-supervised learning process by the visual semantic graph to obtain unmarked association between the image areas and the image area label, thus labeling the unmarked image area. DESCRIPTION OF DRAWING(S) - The drawing shows a flow diagram illustrating a visual semantic graph based image area labeling method. '(Drawing includes non-English language text)'