基于神经网络的男西装图像情感语义识别  

Men's suit image emotional semantic recognition based on BP neural network

作　　者：张海波[1,2] 黄铁军[3] 修毅[1] 赵野军[1] 章江华[4] 

机构地区：[1]北京服装学院计算机信息中心,北京100029 [2]北京服装学院服装材料研究开发与评价北京市重点实验室,北京100029 [3]北京大学信息科学技术学院,北京100871 [4]北京服装学院基础教学部,北京100029

出　　处：《纺织学报》2013年第12期138-143,共6页Journal of Textile Research

基　　金：教育部人文社会科学规划基金资助项目(12YJA760089);北京市自然科学基金资助项目(4102023);服装材料研究开发与评价北京市重点实验室开放课题(2011ZK-03)

摘　　要：为实现基于内容的男西装图像情感语义识别,需要将男西装图像的低层特征映射到情感语义空间。在构建的二维图像情感因子空间和男西装图像视觉特征(十维的亮度—冷暖模糊直方图;七维的饱和度—冷暖模糊直方图+色彩对比度值的综合特征)的基础上,通过机器学习(BP神经网络)实现了男西装图像的低层特征到情感语义因子空间的映射,根据图像低层颜色特征可以自动完成图像情感因子值和情感描述值的计算,并把识别后的新图像数据自动加入图像数据库中。实验结果证明,BP神经网络方法能较好地实现基于内容的男西装图像情感语义的识别。In order to realize the content-based men＇s suit image emotional semantic recognition, the primitive features of men＇s suit image need to be mapped to emotion semantic space. Based on the formed two-dimension image emotional factor space and the primitive features of men＇s suit image （ten-dimension of brightness, warm-cold fuzzy histogram; seven-dimension of saturation, warm-cold fuzzy histogram and color contrast value） in our previous work, the machine learning, BP neural network could help to realize the mapping that from the primitive features of men＇s suit image to the emotion semantic factor space. The primitive features of image color could finish the image emotional factor values and description values calculation automatically, and the new image data could be added after identification to the image emotion database automatically. Experiments have proved that BP neural network method can be a better method to realize the content-based men＇s suit image emotional semantic recognition.

关 键 词：男西装图像 情感语义 颜色特征 BP神经网络 图像识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TS941.712[自动化与计算机技术—计算机科学与技术]