摘要
特征是描述纹理根本属性的一种有效手段,本文结合Gabor滤波器对不同频率和方向的选择特性及自组织特征映射神经网络对特征聚类的适应性和灵活性,提出了一种新的纹理图像分割方法。该方法首先通过Gabor滤波器提取纹理图像的能量特征,然后运用自组织特征映射神经网络进行特征聚类和分类,实现纹理图像的分割。仿真结果证明,该方法能有效地分割出区域特性不同的纹理,且错分率低于共生矩阵和K均值聚类相结合的分割方法。
Feature illustration is an efficient measure to describe the essential property of texture. A new method is proposed for texture segmentation based on energy features by combining the sensitive selectivity of Gabor filters for different frequencies and orientation with the flexibility and adaptability of the self-organizing feature mapping neural network. First, texture energy features are extracted by Gabor filters. Second, the self-organizing neural networks are employed to perform feature clustering, and finally texture segmentation is accomplished by minimum distance classification. Simulations show that this method can efficiently segment multi-texture images into several regions according to their different texture properties, and the segmentation error ratio is lower than the method of combining co-occurrence matrix features extraction with K-means clustering.
出处
《铁道学报》
EI
CAS
CSCD
北大核心
2004年第3期67-70,共4页
Journal of the China Railway Society
基金
兰州交通大学"青蓝"人才工程资助项目
甘肃省高原交通信息工程及控制重点实验室项目
关键词
纹理分割
能量特征聚类
自组织特征映射神经网络
texture segmentation
energy features clustering
self-organizing mapping neural network
