基于自适应定向正交投影分解的图象分割方法

Approach of Image Segmentation Based on Adaptive Oriented Orthogonal Projective Decomposition

将目标和背景分别对应到灰度直方图中的两个高斯分布是进行图象分割的一种常用方法 ,但复杂图象的直方图往往是多峰的 .为了更好地拟合这种复杂图象直方图的多峰特性 ,提出了一种基于自适应定向正交投影分解的图象分割方法 .该方法首先将这种复杂图象的直方图看作是多个高斯分布的叠加 ,并可通过应用自适应定向正交投影分解法来快速准确地确定每个高斯分布的权值、均值和方差 ,进而计算出各相邻高斯分布之间的最优阈值 ,以用于图象分割 .在此基础上 ,又提出了阈值分离度的概念 ,并将其作为选取最终阈值的指标 .应用实例结果表明 ,该方法能够快速有效地实现复杂图象的多阈值分割 .

Looking upon the gray level histogram as a mixture of two Gaussian density functions is a conventional model in the image segmentation, unfortunately the histogram of the complex image often appears a multi-peak feature. In order to get a more accuracy approximation of this kind of histogram, this paper generalizes this model by considering the histogram a mixture of several Gaussian density functions, and employs a new algorithm of Adaptive Oriented Orthogonal Projective Decomposition to handle the mathematical problems involved in this process. In this proposed method, the key parameters of each Gaussian function can be calculated efficiently, which adequately leads to the determination of the optimal thresholds between different neighboring Gaussian functions. A new parameter called the Dividual Ratio of Threshold is introduced and used as the reference for the selection of the final thresholds. Experimental results show that this method can be effectively applied for the multi-threshold segmentation of complex images.