摘要
红外图像往往存在着背景不一致的特点,因此在分割时无法有效地提取出目标。本文针对这一问题提出了一种基于曲面拟合的图像分割方法,以分割非均匀背景下的红外目标。这种方法首先对背景进行光顺限制的曲面拟合,再通过设置一个偏移量来形成阈值曲面。通过研究发现,曲面拟合时的偏离项和光顺项的权重系数比是由图像背景的双拉普拉斯变换以及噪声均方差共同决定的,从而在估计噪声均方差的基础上实现了对权重系数的自适应选取。从仿真结果可以看出,本文提出的曲面拟合分割法在背景去除和目标提取上要优于传统的Ostu法和局部阈值法。
The infrared image often contains undulated and non-consistent background. These characteristics make standard segmentation methods ineffective for extracting target. To solve this problem, a surface fitting segmentation method was put forward. This method includes two steps： surface fitting based on fairing constraint and adding offset to form the threshold surface. Through analysis, it was found out that the ratio of deflection item and fairing item was determined by the results of double-Laplace transform and mean square deviation of noise. So the adaptive ratio could be computed based on estimating mean square deviation of noise. The experimental result shows that the new method has better capability than other often used method in removing background and extracting target.
出处
《光电工程》
EI
CAS
CSCD
北大核心
2006年第7期83-87,共5页
Opto-Electronic Engineering
基金
航空基础科学基金资助项目(04I53067)
高等学校博士学科点专项科研基金资助项目(20020699014)
关键词
图像分割
曲面拟合
阈值曲面
B样条
Image segmentation
Surface fitting
Thresholding surface
B-spline
