摘要
从零件图像的小波分解系数和相对图像边缘像素系数作为零件特征的方法出发,提出了基本概率分配构造和多源零件图像特征识别的方法。首先,对多源零件图像分别进行小波分解,获取零件图像的小波分解系数。对零件图像进行小波多尺度边缘检测,将被检测的零件边缘轮廓图像分成若干个区域并分别统计各区域的相对边缘像素系数。然后,多源零件图像的小波分解系数和零件边缘轮廓图像的相时像素系数作为零件图像的特征并作为神经网络的输入,获取多源零件图像识别的基本概率分配。最后,依据证据理论的合成规则得到零件的识别结果。实验结果表明,基本概率分配构造和多源零件图像特征识别的方法是有效的。
Starting with the coefficients of wavelet transform and the relative image edge pixel co- efficients of part image were considered as features of the part,a method for basic probability assign- ment contribution and feature recognition of multi-source part image was presented.Firstly,the multi-source part image was analyzed based on wavelet transform to obtain the coefficients of wavelet transform,and the edges from part image were detected using wavelet multi-scale edge detection, edge image was divided into several sub-areas,their edge pixels and the relative edge pixel coeffi- cients were counted respectively.Then,the wavelet analyzed coefficients and the relative edge pixel coefficients were considered as features of the part image,which were used as the inputs of a neural network to obtain basic probability assignment.Finally,a part was realized pattern recognition using combination rules of Dempster-Sharer evidential theory.Experimental results show that the pro- posed method is effective for feature recognition of multi-source part image.
出处
《中国机械工程》
EI
CAS
CSCD
北大核心
2006年第S2期58-61,共4页
China Mechanical Engineering
基金
江苏省自然科学基金(05KJB460036)
关键词
小波变换
神经网络
证据理论
图像识别
wavelet transform
neural network
evidential theory
image recognition