【摘要】 图像情感标识就是为图像标注形容词性关键词,以反映用户对该图像的情感或印象.图像的视觉特征以及语义内容是决定用户对该图像情感理解的2项关键因素,而图像内容识别具有较高的不确定性,人类的情感理解也具有很强的主观性,因此采用Dempster-Shafer证据理论能较好实现图像视觉特征及语义内容到图像情感标识的不确定性推理.考虑到图像内容识别的不确定性,研究中按一定比例扩大了图像语义内容对各情感因子的不确定性区间,并构建了一个原型系统对自然风景图像进行自动标识.实验表明Dempster-Shafer证据理论在处理情感标识上是很有效的,而调整不确定性区间有助于进一步提高标识准确率.更多还原

【Abstract】 Affective image annotation involves labeling an image with adjectives,so that those labels reflect the user’s emotional understanding of the image.The low-level visual features and the image semantic content are two decisive factors in the user’s emotional understanding of an image,while image content recognition is highly uncertain and affective understanding is strongly subjective.In the following study,the Dempster-Shafer theory was applied to represent the visual image characteristics and to model the uncertainty reasoning from those decisive factors to affective understanding.In response to the semantic recognition error,the uncertainty range of image contents to each affective factor was enlarged and a prototype affective annotation system was built to automatically label natural scenic images.Experimental results show that the Dempster-Shafer theory is promising for ambiguous annotation,and enlarging the uncertainty range is helpful for improving annotation precision.更多还原