轮廓点分布直方图CPDH(Contours Points Distribution Histogram)是一种形状描述子,但它对微小形变比较敏感且在大数据集下的检索效果不佳.提出基于协同传递机制的半监督学习框架Co-transduction与CPDH相结合的算法.通过给定CPDH的相似度度量和另一种描述符的度量结果,对一幅查询图像,利用其中一种度量准则迭代检索出与查询图像最相似的目标形状将其标记.用另一种相似性度量重新检索并排序已标记的形状,反之亦然.该改进算法较原始CPDH在大数据集下(MPEG-7)的检索性能更优,检索精确率达到86％,比原算法提高约10％.%Contour points distribution histogram (CPDH) is a shape descriptor,but it is sensitive to small deformations and under the big data set under the retrieval effect is not good.A Semi-supervised learning framework based on cooperative transmission mechanism co-transduction combined with CPDH algorithm was proposed.Given the similarity measure of CPDH and the measurement result of another descriptor,for a query image,the most similar target shape to the query image was retrieved iteratively using one of the metric criteria and labelled.Then another similarity measure was used to retrieve and sorted the marked shapes,and vice versa.Compared with the original CPDH,this improved algorithm has better retrieval performance under the big data set (MPEG-7),and the retrieval accuracy is 86％,which is about 10％ higher than the original algorithm.
展开▼
