激光夜视图像分型分割算法研究

为解决以往采用关联规则挖掘算法对图像进行分割时,对于夜视图像中灰暗区域中颜色特征以及前景/背景特征的采集能力差,不能有效判定图像的多尺度分型特征,分型分割效果差的问题。研究激光夜视图像分型分割算法。先利用计盒维数估计方法计算激光夜视图像分型维数尺度,通过分型维数尺度获取激光夜视图像的多尺度分型特征值,将利用多尺度分型特征值获取的多尺度分型特征约束与图像颜色约束相结合获取多尺度分型特征数据项,融合该多尺度分型特征数据项与通过图像中相邻区域顶点颜色距离获取的光滑项,并加入自适应比重系数获取能量函数,利用最大流/最小割算法求解能量函数最小值,实现激光夜视图像的分割。实验结果表明,该算法可准确分割激光夜视图像中人物目标特征,分割10幅激光夜视图像准确率以及均匀性测度平均值均在95%以上。

Research on the segmentation algorithm of laser night vision image segmentation

In order to solve the problem of poor acquisition ability of color features and foreground/background features in dark areas of night vision images,multi-scale segmentation features of images cannot be effectively determined and the segmentation effect is poor when the image is segmented by association rule mining algorithm in the past. The algorithm of laser night vision image segmentation is studied. First the box-counting dimension estimation method is used to calculate laser night vision image segmentation dimension. The multi-scale segmentation feature values of laser night vision image captured by the segmentation dimension scale will use multi-scale segmentation feature constraints obtained by multi-scale segmentation feature values to combine with image color constraints for obtaining multi-scale segmentation feature data items. The multi-scale segmentation feature data items and smooth items obtained by vertex color distance of adjacent area in image are fused,and the adaptive weighting coefficients are added to obtain energy function. Using maximum flow/minimum cut algorithm to solve the energy function minimum to realize laser night vision image segmentation. The experimental results show that the algorithm can segment the character target features in laser night vision image accurately,and the average accuracy and uniformity of every 10 laser night vision images are above 95%.