结合各向异性扩散和分水岭算法的图像分割  

Image Segmentation Based on Anisotropic Diffusion and Watershed Algorithm

作　　者：陈世文[1] CHEN Shiwen(College of Computer Science,Jiaying University,Meizhou Guangdong 514015,China)

机构地区：[1]嘉应学院计算机学院,广东梅州514015

出　　处：《传感技术学报》2023年第10期1622-1627,共6页Chinese Journal of Sensors and Actuators

基　　金：嘉应学院项目(2016KJZ03)。

摘　　要：传统分水岭算法受到噪声干扰时容易出现过分割现象,为了抑制噪声的同时尽可能多地保留住图像边缘信息,提出了一种基于各向异性扩散的分水岭分割算法。首先对原始图像进行滤波操作,传统P-M算子通常人为设定固定边界阈值,容易丢失细节信息,应用梯度模的变化设定阈值并连结图像结构张量形成一个扩散函数,边缘处沿切线方向扩散易于保留边缘细节,平坦处具有各向同性易于平滑噪声,这样保证了良好的分水岭结构。其次对图像的梯度信息进行计算,为了使梯度信息得到补偿,采用数学形态学的开闭运算对图像梯度信息进行处理。然后运用形态学极小值标定方法标记处理后的图像局部极小值,最后用分水岭算法对图像进行分割。实验对无噪声图像和加噪声图像进行分割,结果表明该方法具有良好的分割效果,尤其对噪声图像有较好的鲁棒性。Traditional watershed algorithm is easy to cause over segmentation.In order to preserve edge information and suppress noise,a watershed segmentation algorithm based on anisotropic diffusion is proposed.Firstly,the original image is filtered.The traditional P-M operator adopts a fixed boundary threshold,which is easy to lose detail information.The threshold is set by the change of gradient modulus in combination with the image structure tensor to construct a diffusion function.The edge details are preserved by diffusion at the edge,and the noise is smoothed at the flat part,so that it has a good watershed structure.Secondly,gradient information of the image is calculated.In order to compensate the gradient information,the opening and closing operation of mathematical morphology is used to process the image gradient information.Then,the morphological minimum calibration method is adopted to mark the local minimum of the processed image,and finally,the watershed algorithm is applied to segment the image.Experimental results show that the proposed method has good segmentation effect,especially good robustness to noisy images.

关 键 词：图像分割 分水岭算法 各向异性扩散 P-M模型 标记提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]