摘要
针对无指针式表盘的数字判读问题,提出一种基于Zernike矩和粗集预处理的神经网络数字识别方法。该方法首先利用Zernike矩的旋转不变性特征提取数字图像特征,再对所提取的Zernike矩进行基于粗集的特征约简,约简后的信息输入到训练好的神经网络进行识别。通过实际的表盘分割截取的带旋转的数字识别中试验,结果表明该方法具有识别率高,速度快的特点,具有较高的实时价值。
Aiming at number recognition of no-pointer dials, a new arithmetic based on neutral network with Zernike moments and rough sets was put forward. This method takes Zernike moments as image features for their rotate-invariety, then rough sets theory was used to reduce features, resuhs can be taken as input information of neutral network. Experiments resuhs indicated this method had many advantages, such as precise and effective.
出处
《微计算机信息》
北大核心
2005年第12Z期172-173,155,共3页
Control & Automation
基金
国防预研基金资助项目
