自适应脉冲耦合神经网络在图像处理中应用

Study on Self-adaptive Pulse Coupled Neural Network and Its Application in Fields of Image Processing

尽管Johnson提出的PCNN模型具有强大的图像处理功能,以时间序列进行特征提取时具有旋转、尺度、平移、扭曲不变性,可实践中发现依然存在着不足,特别对图像亮度、对比度比较敏感。添加了误差反向传播(Error Back Propagation,EBP)学习准则的自适应脉冲耦合神经网络模型能自适应设定模型参数,是脉冲耦合神经网络模型研究的主要内容。特别地,应用这种自适应模型进行特征提取时,能弥补原来PCNN模型对亮度、对比度敏感的缺陷,而且具有一定的泛化能力,有效克服了亮度、对比度对图像识别精度的影响。

The standard Pulse Coupled Neural Networks （PCNN） has been widely used in the image processing, however, it is hard to set plenty parameters of PCNN efficiently which limited its capability for image processing. Based on the learning rules, PCNN was optimized through running its parameters adaptively. A gradient descent algorithm was adopted to search parameters which could reduce the error between the desired output and the actual output gradually according to the least mean square principle. The traditional PCNN model is used to image feature extraction, its output features are rotation, scale and shift invariant, but it is sensitive to illumination, therefore the adaptive parameters PCNN is used for image feature extraction when the stimuli＇s illumination （intensity or contrast） is varied. The results are shown that the application efficiency of feature extraction is improved.