基于自组织神经网络的彩色图像自适应聚类分割

Color Image Self-adapting Clustering Segmentation Based on Self-organizing Feature Map Network

针对一般聚类分割算法对于色彩丰富、背景复杂的图像容易造成聚类重叠,引起像素错误分类的缺点,提出一种新的基于自组织特征映射神经网络的彩色图像分割方法.首先利用各像素的RGB值作为输入样本对网络进行训练;然后根据竞争层特征映射点的密度分布图,利用自组织映射分析的方法,确定图像颜色的聚类数和聚类中心;最后利用距离竞争取胜的原则处理每个像素,从而实现彩色图像的区域分割.通过实例验证,该方法能够较好地完成彩色图像的自适应聚类分割,处理效果良好.

The traditional clustering-segmentation algorithms can produce clustering overlap easily, and result in error classification of image pixels for colorful images with complicated background. To solve this problem, a new method of color image segmentation based on self-organizing feature map （SOFM） neural network is presented. The first step is training the network with RGB values of the image elements and then getting the density distribution graph of whole feature mapping points. The next step is by means of self-organizing map analysis （SOMA）, finding out the clustering number and its centers automatically. Finally, every element is calculated and classified according to the distance competition. Examples show that the method is effective on color image self-adapting clustering and segmentation.