现代成像制导中所使用的红外图像往往存在着噪声大、目标-背景间灰度差较小、边缘较模糊的特点.这些特点会增加边缘提取的难度,因此必须建立更有效的红外图像边缘提取算法以满足需要.针对这些问题,以噪声Gauss分布模型和噪声特征为基础,建立了新型统计学意义下的红外图像边缘检测法.通过对此方法的概率模型进行分析,可以证明在有较大噪声的情况下,只要边缘处的差分值大于一定的值,就能以较大的概率提取出图像边缘.通过在不同情况下与梯度法的抑噪能力进行对比和分析发现,统计边缘提取法的噪声抑制能力要高于梯度法.在与Sobel模板算子法的红外图像边缘检测结果进行仿真和对比后发现,统计法能对红外图像的目标边缘检测取得良好的结果,并且算法具有快速简单的优点.

Characteristics of IR images mostly include large noise, small distinction betweens target and background, blurry edge and so on. These characteristics make existent methods ineffective and require more effective method. Considering the random noise often accords with Gauss distribution, a statistic method based on noise probability distribution and noise characteristics has been put forward to detect image edge. By analyzing the probability model of noise,it is proved that if the difference of neighboring dots is larger than a determinate value, the edge of infrared image can be detected in large probability. By comparing the infection of noise for statistic method and gradient method in different conditions, it is found that the noise restrain capacity of statistic method is higher than that of gradient method. Comparison and analysis are made between the statistic method and the often used Sobel algorithms. The analyses indicate that good results can be obtained when this statistic method is used to detect edge of infrared image and this method is simple and fast.