基于差别特征学习的神经网络图形识别

Recognition of Graphics Using Neural Networks Based on Difference Feature Patterns

根据视觉识别的差别特征分辨特性，定义了差别特征模式（ＤＦＰ），对前向人工神经网络进行了改进，提出了基于差别特征模式的学习方法．针对视觉识别的大小不变性、位置无关性、轮廓与实心等价性以及变形有噪时的相对稳定性，设计了前向网络，通过调节注视中心来识别多个物体．实验结果证实，该神经网络对模式具有较好的识别稳定性．

Thisarticle discussed the vision and itsrecognition processand found outthathum an recognizes objects not by their isolated features,but by their m ain difference features w hich people get through contrasting them .According to the principle,the difference feature patterns(DFP) w ere defined,based on w hich a new learning m ethod w aspresented and BPneuralnetw orksim proved.Considering the featuresof vision recognition,such assize unchangeablefeature,aBPnetw ork w asdesigned which recognizesm ultiple objects by using focusofattention.The experim entprovesthe recognition stability ofthe netw ork.