提出了一种基于函数联接的感知器神经网络的纹理分类方法.它采用高斯2马尔柯夫随机场模型(GM RF)对纹理进行描述,模型参数即为纹理特征,参数估计采用最小平方误差方法获得.将估计参数作为表达纹理的特征向量,用感知器网络对特征进行分类,并且采用函数联接的方式解决线性不可分问题.对纹理图象进行的实验表明,采用这种方法能够提高学习速度,简化计算过程,并取得较好的纹理分类效果.

-Based on the function connected perceptron neural network texture classification method. It uses 2 Gaussian Markov Random Field Model (GM RF) to describe the texture, the model parameters is the texture feature, parameter estimation using least squares error obtained. the estimated parameters as the expression of texture feature vector, using the characteristics of sensor networks for classification, and the use of function to resolve connection problems can not be separated from linear. of texture images of the experiments show that this approach can enhance the learning speed, to simplify the calculation process and obtain a better effect of texture classification.