摘要
为提高图像标注的准确率,提出了根据图像复杂度采用相应视觉特征表示机制的方法。对待标注的简单图像,直接提取其全局特征信息;对待标注的复杂图像,采用分割、过滤技术处理后,提取每个分割块的局部特征信息。通过训练图像数据学习了贝叶斯分类模型,用该分类模型对未标注图像进行标注。实例验证了该方法比采用单一视觉特征表示机制方法具有更好的标注效果。
To improve the accuracy of image annotation,corresponding visual feature representation mechanisms based on image complexity is proposed.Global feature of unmarked simple image was extracted directly.The technology of segmentation and filtration was applied to unmarked complex image,then local feature of each block is extracted.Bayesian classification model witch can annotate the unlabeled image was learnt through training image data,Finally the experiment results showed that the proposed method could achieve better annotation effect than that using only one visual express mechanism.
出处
《计算机工程与设计》
CSCD
北大核心
2014年第6期2100-2103,共4页
Computer Engineering and Design
基金
国家自然科学基金项目(61170145)
教育部高等学校博士点专项基金项目(20113704110001)
山东省自然科学基金
科技攻关计划基金项目(ZR2010FM021
2008B0026
2010G0020115)
关键词
贝叶斯分类器
颜色特征
纹理特征
图像分割
图像聚类
图像标注
Bayesian classifier
color feature
texture feature
image segmentation
image clustering
image annotation
