摘要
为了快速、准确地标注大型图像数据集中的图片,提出了一种利用图像分割和基于kNN(k-nearest neighbor)图的半监督学习来标注图像的算法.该算法先将图像分割为若干个局部区域,使用局部敏感的哈希表来构建图像局部区域的kNN图,并基于图像局部区域的kNN图来构建原始图像的kNN图,利用基于图的半监督标签传递算法来标注未标注的图像.在具有269 648张图像的大型图像数据集NUS-WIDE和具有5 000张图像的Corel数据集上的实验结果表明,该算法能获得较快的标注速度和标注精度.
In order to annotate the images in large image datasets quickly and accurately, an algorithm using image segmentation and kNN （k-nearest neighbor） graph-based semi-supervised learning for image annotation was proposed. The images were segmented into several local regions with the proposed algorithm, and the kNN graph of local regions was established with the locally sensitive hash table. In addition, the kNN graph of original images was constructed based on the kNN graph of local regions of images. The graph-based semi-supervised label propagation algorithm was used to annotate the images which were not annotated. The experimental results of both large image dataset NUS-WIDE with 269 648 images and Corel dataset with 5 000 images show that the proposed algorithm can obtain higher annotation speed and annotation accuracy.
出处
《沈阳工业大学学报》
EI
CAS
北大核心
2013年第4期438-444,共7页
Journal of Shenyang University of Technology
基金
国家自然科学基金资助项目(60973105)
国家863先进制造领域重点项目(2008AA04A120)
关键词
图像分割
半监督学习
图像标注
哈希表
kNN图
标签传递
聚类算法
方向梯度直方图
image segmentation
semi-supervised learning
image annotation
hash table
kNN graph
label propagation
clustering algorithm
histogram of oriented gradient
