Image semantic auto-annotation method based on neighborhood and distance metric learning



The invention provides an image semantic auto-annotation method based on neighborhood and distance metric learning. The method includes: by introducing a transformation matrix, randomly selecting any two images from a training set to obtain a distance metric; computing prior probability of each annotation word, acquiring a neighborhood of each image in the training set, recording occurring times of the annotation word in the training set and absence times thereof, and computing conditional probability; acquiring a neighborhood of each image in a testing set, and acquiring a standard word vector by computing an image coefficient and outputting the same. The method has the advantages that the number of annotation words need not be determined in advance, the intelligence level is higher than that of the prior art, annotation results are more accurate, the neighborhoods of the images are all acquired through distances obtained by learning, and precision is higher.




Download Full PDF Version (Non-Commercial Use)

Patent Citations (1)

    Publication numberPublication dateAssigneeTitle
    CN-104008177-AAugust 27, 2014华中师范大学Method and system for rule base structure optimization and generation facing image semantic annotation

NO-Patent Citations (4)

    Image annotation using high order statistics in non-Euclidean spaces;Songhao Zhu 等;《Elsevier》;20130920;第1342-1348页
    Image Annotation using Metric Learning in Semantic Neighbourhoods;Yashaswi Verma 等;《European Conference on Computer Vision》;20121031;第1-15页
    一种适合弱标签数据集的图像语义标注方法;田枫 等;《软件学报》;20130503;第2405-2418页
    基于视觉语言模型的图像语义挖掘研究;金聪 等;《图书情报工作》;20130331;第57卷(第5期);第120-123页

Cited By (0)

    Publication numberPublication dateAssigneeTitle