摘要
通过融合图像中不同模态的信息并利用少量带标记的图像进行半监督距离学习,来对图像进行聚类。首先,提取彩色图像中RGB颜色空间的直方图信息、纹理信息,并采用SIFT算法提取Bag of Words来重新表达图像,从而基于图像的颜色特征、纹理特征以及语义特征,建立图像的多模态表达机制,将原始图像投射到表达空间;然后,利用少量标记的图像,通过半监督距离学习,获得图像在多模态信息空间的相似性度量;最后,通过半监督聚类方法,实现图像分组,在多个图像数据库中验证提出的方法的有效性。
The project clustered images by fusing the different model information in the images and taking advantage of a small amount of labeled images for semi-supervised distance lesming.First,we extracted histogram information of the RGB color space,texture information in the color images,and Bag of Words by using the SIFT algorithm to re-express the image,thus establishing the multi-modal express mechanism of images based on the image's color,texture and semantic features to project the original image onto the space to express.Then,using a small amount of the marked image,we obtained a similarity measure in multi-modal information space of images through the semi-supervised distance learning.Finally,we realized grouping images through the semi-supervised clustering method and verified the validity of the proposed method in the plurality of images in the database as well.
出处
《计算机科学》
CSCD
北大核心
2014年第3期41-45,共5页
Computer Science
基金
国家优秀青年科学基金(61222210)资助
关键词
半监督
距离学习
多模态
图像聚类
Semi-supervise
Distance learning
Multi-modal
Image clustering