基于规一化割的声呐图像谱抠图分割

针对目前谱图理论应用于声呐图像分割时,效果不够理想的问题,提出一种结合Ncut和谱抠图的声呐图像分割方法.该方法首先通过形态学变换对声呐图像进行预处理,降低复杂背景对分割结果的影响；其次,引入数字抠图技术,通过改变Ncut算法中的拉普拉斯公式得到用于分割的图像透明度估计；最后,通过透明度处理得到最终的声呐图像分割结果.仿真实验证明了所提算法的有效性,与传统的谱图分割方法相比,没有对声呐图像的背景进行分割,并准确提取出了目标区域,得到了更理想且更细致的声呐图像目标分割结果,有利于后期识别.

Sonar image spectral matting segmentation based on normalized cut

When applying spectral graph theory to sonar image segmentation,the results are often not ideal.To solve the problem,a sonar image segmentation method combining normalized cut and spectral matting was presented.Firstly,morphological transformation was used for sonar image pretreatment to reduce the effect of a complex background.Secondly,digital matting techniques were introduced and the Laplace equation of the normalized cut algorithm was changed to obtain the image transparency estimation for segmentation.Finally,the sonar image segmentation results could be obtained by transparency processing.The simulation experiment shows the effectiveness of the proposed algorithm.Compared to the traditional spectral segmentation method,this algorithm does not segment the background of a sonar image,and can extract the target region accurately,obtaining better and more detailed sonar image object segmentation results.This improvement is also conducive to later identification.