摘要
提出一个基于EM迭代的非监督图像多标签区域标定算法,它能够非常有效地将基于全图的标签自动标定到图像的对应局部区域上。首先对所有图像进行SIFT特征点的密集采样,然后对所有的SIFT特征点进行K-m eans聚类,获得词典,再构造EM迭代过程计算出每幅图像中每个标签对每个存在WORD的置信度,最后选择那些置信度较高的WORD,确定每幅图像中每个标签置信度最高的对应区域。实验表明,在样本数据充分的情况下,该算法在解决非监督自动标定、标签表观的多样性以及多标签等问题上都取得了不错的效果。
In this paper,we propose an EM iteration-based unsupervised regions annotation algorithm for image multi-labels,which can automatically reassign the manually annotated labels from the image-level to their corresponding local semantic regions effectively.First,we extract SIFT feature points by dense sampling from all the images in the dataset and create the dictionary by K-means clustering on all the extracted SIFT feature points.Then an EM iterative process is constructed to calculate the confidence of each label with regard to each existed WORD in each image.Finally,we choose those WORDs with highest confidence to determine every corresponding region in each image having the label of highest confidence.Experiments demonstrate the encouraging performance of our proposed algorithm in solving the problems of unsupervised automatic annotation,appearance diversity of labels and multi labels,etc.on the premise of sufficient samples data.
出处
《计算机应用与软件》
CSCD
北大核心
2012年第2期5-8,26,共5页
Computer Applications and Software
基金
国家自然科学基金(60875003)
上海市科委重大专项(10dz1500104)