自然风景图像情感标识方法研究  

Research on Affective Annotation for Natural Scene Images

作　　者：高彦宇[1] 王新平 尹怡欣[1] 

机构地区：[1]北京科技大学信息工程学院,北京1000083

出　　处：《小型微型计算机系统》2011年第4期767-771,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60374032)资助

摘　　要：图像情感标识就是为图像标注情感类关键词以反映用户对该图像的情感或印象.以自然风景图像为对象,提出一种由视觉特征提取、视觉-情感关系构建、情感标识推导三部分组成的图像情感标识方案.首先从情感认知角度提取自然风景图像的颜色、纹理等视觉特征,然后采用多维尺度技术挖掘图像视觉特征和人类主观感知之间的深层联系,最后采用贝叶斯决策模型根据图像视觉特征推导出其情感描述.对部分风景图像进行自动标识的结果表明贝叶斯决策模型在处理情感标识这类不确定性推理问题上是很有效的.Affective image annotation is to label an image with affective adjectives, so that those labels could reflect user＇s emotion or impression to the image. In this paper, we presented an affective annotation scheme for natural scene images, which consisted of three parts： extracting visual features, building visual-emotion relationship, and inferring affective perception. In the In：st part, low- level visual features such as color and texture features were extracted from perceptual viewpoints for all images. In the second part, the multidimensional scaling technique was applied to discover the relationship between image visual features and human＇s emotion. In the third part, a Bayesian decision model was built to characterize an image with 5 affective scores based on the visual-emotion relationship. A prototype system has been developed, the experimental results of which show that the Bayesian decision models are promising in affective image annotation.

关 键 词：图像情感标识 多维尺度技术 贝叶斯决策模型 语义差分 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]