【摘要】 综合利用红外图像灰度特征与可见光图像灰度以及分维数方差特征,提出了一种复杂背景下多目标分割算法。首先通过最大熵阈值分割出红外图像的潜在目标区域,记录其质心位置及形状大小并对应到可见光图像中。再提取可见光图像的分形维数,利用其方差特征增强目标奇异性,排除背景奇异区域干扰,并对记录的目标区域进行初判决,得到真实目标质心处的分维数方差。然后将分维数图划分为与已知目标大小接近的区域块,搜索并标记具有相近分维数方差的所有区域块。最后在所标记及其相邻的区域块中精确分割出全部目标。对大量实际图片进行的仿真实验证明,该分割算法可以有效地进行多目标分割,并较好地保留目标形状特征。更多还原

【Abstract】 Synthetically utilizing features coming from infrared images and visible-light images, a novel segmentation algorithm is proposed to deal with multi-target in complicated backgrounds. Firstly, a threshold is determined to segment infrared images by Maximum Entropy Model and information of regions containing potential targets is recorded. Secondly, Fractal Dimension(FD), extracted from intensity distribution of visible-light images, is used to calculate statistical features. Thirdly, regions in FD distribution images corresponding to the recorded parts in infrared images are judged and classified into targets or backgrounds. Then dividing the FD images into regions with a size close to the targets, and those having similar variances are selected out. In the end, all targets could be segmented from complex backgrounds. A large number of simulative experiments on real scene prove the validity and reliability of the scheme.更多还原