【摘要】 提出了一种基于不变矩特征和神经网络的医学图像识别模型·所设计的识别模型包括不变矩特征提取、不变矩矢量标准化、模糊化预处理、BP网络与竞争选择·利用不变矩方法提取医学图像的特征矢量,能有效检测出具有平移、旋转和比例变化的图像,利用神经网络作为分类器对提取的特征矢量分类,使用模糊化的方法先对输入特征数据做预处理再进行识别,每一个图像模式归属于某一类是以0到1的数字代表其归属程度·实验结果验证了模型的有效性,训练好的网络有很好的分类能力·更多还原

【Abstract】 A medical image recognition method based on moment invariant feature and neural networks is proposed, including the moment invariant feature extraction, moment invariant vector standardization, fuzzy preprocessing, BP net and competition selection. The feature vector of medical images, as extracted by the method of moment invariant, can effectively recognize the images characterized by translation, rotation and scaling invariants. Utilizing neural networks for classification, the extracted feature vector is classified. By use of fuzzy method, the feature datainput is preprocessed then recognized. Thus, the attribution of each and every image pattern is supposed to be expressed by a number from 0 to 1 to indicate how an image pattern is attributed to a class/sort. Experiment results demonstrated that the method is effective, and the net possesses high classing ability if trained up.更多还原