可见光图像背景起伏的平稳性和相关性分析

Stationary characterization and correlation of visible image background clutter

提出了单位根假设检验可见光图像背景起伏平稳性的方法,理论上证明了时空域变化的广义平稳性。该方法采用ADF假设检验,从时域和空域分析背景灰度的变化,利用线性回归的方法进行协整性检验,从而得出结论。在平稳的基础上,使用相关长度刻划背景起伏的时空域相关性,得到两个结论。(1)一维指数模型的时间相关长度分析表明背景起伏具有强时相关性。并且这种强相关性集中体现在像素点灰度均值和相邻的有限帧(1～2帧)。因此提出低阶AR模型(1～2阶)对这种特性进行描述,得到的模型参数可用于图像背景估计。(2)空间相关长度分析证实了背景起伏具有强的空域相关性。并且图像可以分割成若干个不相关的区域。总的来说,背景起伏具有时空域的双重平稳性和强相关性。

Quantitative analysis visible image background clutter stationary method via the unit root test was presented. It is proved in theory that the clutter is generalized temporal-spatial stationary. The method used the ADF test to analyze the clutter temporal-spatial variation, and used line-regress-method to test cointegration, then made that conclusion. Based on the stationary, the correlative length was applied to describe its temporal-spatial correlation, and then two conclusions were made. （1）The temporal correlative length is proposed based on 1D exponent model and it shows that the clutter is highly temporal correlative. Furthermore the correlation is incarnated mostly by the sequence mean and some neighbors finite frames （1~2 frames）. Thus low order AR model （1-2 order） is presented to describe the temporal correlation. And these model coefficients can be used to estimate image background. （2）The spatial correlative length analysis proves the clutter highly spatial correlative. Furthermore, images can be segmented to some areas uncorrelative with each other. In a word, the clutter is temporal-spatial stationary and correlative highly, which are important to background suppression and detecting targets.