基于曲率相似性的SAS图像线状目标提取

Linear Target Extraction from SAS Image Based on Curvature Similarity

针对合成孔径声呐图像中常见的线状目标的提取,提出了一种基于曲率相似性的最小代价路径提取方法,利用局部曲率相似性来减少噪声对最小路径方法的严重干扰,最终实现目标的自动提取。算法首先针对水下成像中的低对比度、模糊和噪声污染等特性,采用Retinex方法对图像进行亮度均衡和增强处理;然后将图像的像素定义为图的节点,相邻的像素连结为边,将图像表示为一个格状图,对格状图中的最小代价路径进行搜索;最后为了有效减小噪声与干扰的影响,利用线状目标局部路径曲率的相似性作为代价函数的加权约束因子,从而得到目标的完整路径。实验结果表明,本文方法可有效的从合成孔径声呐图像中提取出线状目标。

Aiming at the extraction of common linear targets in synthetic aperture sonar images,this paper proposes a minimum cost path extraction method based on curvature similarity,which uses local curvature similarity to reduce the serious interference of noise on the minimum path method,and finally achieves the goal Automatic extraction.The algorithm first uses the retinex method to perform brightness equalization and enhancement processing for the characteristics of low contrast,blur,and noise pollution in underwater imaging;then,the pixels of the image are defined as nodes of the graph,and adjacent pixels are connected as edges.The image is represented as a trellis graph to search for the least cost path in the trellis graph;finally,in order to effectively reduce the influence of noise and interference,the similarity of the local path curvature of the linear target is used as the weighted constraint factor of the cost function,thus get the full path of the target.Experimental results show that this method can effectively extract linear targets from synthetic aperture sonar images.