【作者】 吴小俊； 张媛媛； 王士同； 於东军； 郑宇杰； 杨静宇；

【Author】 WU Xiao-jun~1 ZHANG Yuan-yuan~2 WANG Shi-tong~1 YU Dong-jun~3 ZHENG Yu-Jie~3 YANG Jing-yu~3 (1.School of Information Engineering,Jiangnan University,Wuxi 214036,China; 2.School of Electronics and Information,Jiangsu University of Science and Technology,Zhenjiang 212003,China; 3.School of Information,Nanjing University of Science & Technology,Nanjing 210094,China)

【机构】 江南大学信息工程学院； 江苏科技大学电子信息学院； 南京理工大学信息学院；

【摘要】 对基于模糊神经网络的人脸图像分类器进行研究。将多输入单输出模糊推理系统改造成多输入多输出的模糊神经分类器,并提出了一种改进的模糊神经分类器,改进模型的计算量明显减少。在将模糊规则库与训练样本集对应的基础上提出了一种模糊隶属函数参数的初始化方法。该初始化方法的优点在于它充分利用了训练样本所包含的鉴别信息。在 ORL 人脸的原始图像空间中用上述方法设计分类器,获得了较好的实验结果。更多还原

【Abstract】 A study on classifier of face images based on fuzzy neural network(FNN)was presen- ted.A MISO(multiple-input sigle output)fuzzy inference system was modified into a MIMO (multiple-input multiple output)fuzzy-neuro classifier.An improved fuzzy-neuro classifier was proposed which reduced the computational load apparently.An initialization method of parame- ters of fuzzy membership functions was suggested based on the correspondence between fuzzy rule base and training set.The advantage of the new initialization method lied in fully utilizing the dis- criminant information contained in the training set.Good experiments results were obtained using the fuzzy-neuro classifier designed in the original image space of ORL(olivertti research lab)ima- ges.更多还原