【摘要】 目的:针对目前计算机辅助诊断的需要以及较难实现语义检索和较少涉及医学彩色图像分类的问题,设计医学内窥镜图像语义标注分类器。方法:基于C#/C++编程语言、Windows 7、.NET、Visual-Studio(VS2008)平台,通过CBIR特征提取,主成分分析(principal component analysis,PCA)降维处理,然后通过BP神经网络训练进行分类,开发应用分类器。结果:分类器对7类医学彩色内窥镜图像分类准确性达到80%,而训练时间只有几秒或几十毫秒。结论:主成分分析和BP神经网络的结合使用,克服了低级特征和高级语义间的语义鸿沟,降维处理大大减小了系统存储量,提高了训练速度,获得了更好的标注结果。更多还原

【Abstract】 Objective To design endoscope image semantic annotation classifier aimed at the needs of computer-aided diagnosis,as well as the problems that it was more difficult to achieve semantic retrieval and less involved in medical color image classification.Methods The classifier was developped based on C#/C++ programming language,Windows 7,.NET and Visual-Studio(VS2008) platform,through the CBIR feature extraction,principal component analysis(PCA) to reduce the dimensions,and then BP neural network training classification.Results The classification accuracy of seven types of medical endoscopic color image was above 80%,while the training time was only a few seconds or tens of milliseconds.Conclusion The combination of principal component analysis and BP neural network overcomes the considerable gap between the low-level features and advanced semantic.The dimensions reducing significantly reduces the system memory,improves the training speed and achieves a better annotation results.更多还原