基于不可分小波与标记分水岭的图像分割  

Image segmentation based on non-separable wavelet and marked-watershed

作　　者：刘斌 程晓玲 LIU Bin;CHENG Xiaoling(School of Computer and Information Engineering,Hubei University,Wuhan 430062,China)

机构地区：[1]湖北大学计算机与信息工程学院,湖北武汉430062

出　　处：《现代电子技术》2020年第15期42-46,50,共6页Modern Electronics Technique

基　　金：国家自然科学基金面上项目(61471160);湖北省自然科学基金重点项目(2012FFA053)。

摘　　要：针对传统的分水岭算法过分割的问题,提出一种结合不可分小波与标记分水岭的图像分割算法。根据二维不可分小波理论,文中利用伸缩矩阵为[2,0;0,2]的不可分小波的构造方法构造了四通道的对称滤波器组,并把它应用到标记分水岭的图像分割中。利用该滤波器组对原图进行不可分小波变换,对分解出的低频子图进行标记分水岭分割;然后对利用标记矩阵进行区域平均后的低频子图进行小波逆变换,得到区域平滑图;最后,对最终的平滑图使用标记分水岭算法,得到分割结果。实验结果表明,该方法可以有效地解决分水岭过分割问题,且较好地保留了图像目标的轮廓信息。In order to solve the over-segmentation problem of the traditional watershed algorithm,a new image segmentation method combining non-separable wavelet and marked-watershed is proposed.According to the two-dimensional non-separable wavelet theory,a four-channel symmetrical filter bank is constructed by using the method of non-separable wavelet,whose dilation matrix is[2,0;0,2],and it is adopted in the image segmentation of marked-watershed method.The filter bank is used to perform non-separable wavelet transform on the original image,the marked watershed segmentation of the decomposed lowfrequency sub-image is conducted,and then the wavelet inverse transformation of low-frequency sub-image after regional average by the mark matrix is performed to obtain a regional smooth image.Finally,the smooth image is processed with marked-watershed algorithm to get the segmentation result.The experimental result shows that the proposed method can effectively solve the over-segmentation problem of the traditional watershed algorithm,and better preserve the contour information of the image.

关 键 词：图像分割 不可分小波 标记分水岭 区域平滑图 对称滤波器组 小波逆变换 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391.4[电子电信—信息与通信工程]