高斯尺度空间下铁谱图像自顶而下的分割方法  

Top-down segmentation method of ferrrographic image in Gaussian scale space

作　　者：宋佳声 王永坚 周海峰 SONG Jiasheng;WANG Yongjian;ZHOU Haifeng(Fujian Provincial Key Laboratory of Naval Architecture and Ocean Engineering,School of Marine Engineering,Jimei University,Xiamen 361021,China)

机构地区：[1]集美大学轮机工程学院,福建省船舶与海洋工程重点实验室,福建厦门361021

出　　处：《厦门大学学报（自然科学版）》2020年第6期1047-1052,共6页Journal of Xiamen University：Natural Science

基　　金：福建省自然科学基金(2020J01686)。

摘　　要：在铁谱图像阈值分割的结果中,图像的前景往往含有大量的非目标磨粒区域.为了提高阈值分割算法的准确性,降低误检率,提出了一种铁谱图像在高斯尺度空间下自顶而下的分割算法.采用卷积和下采样技术,构建由多组分层的平滑铁谱图像所组成的高斯尺度空间,其尺度按照一定规律由细到粗变化;对尺度空间的每一层图像进行独立的阈值分割,不同尺度下的图像采用不同的阈值,在尺度空间中形成了多阈值的分割结果;通过逻辑运算在尺度空间中将这些分割结果自顶而下逐步精细化.针对两类铁谱图像的分割过程表明,从大尺度到小尺度的自顶而下分割过程不仅抑制了绝大部分干扰,而且能够由粗到细逐步找回在小尺度分割结果中所得到的目标细节.最终的分割实验结果表明,该算法能够将传统阈值算法的准确性提高14个百分点,误检率降低15个百分点.Thresholding segmentation results of ferrographic images usually contain abundant uninterested regions.To improve the accuracy and reduce the false alarm rate,we propose a top-down segmentation method in the Gaussian scale space.First,the Gaussian scale space,built by convolution and down sampling,is organized in groups and several levels of smoothed ferrographic images with the scale changing from fine to coarse.Second,thresholding segmentation is implemented in all smoothed images in the scale space.These segmentation thresholds differ from one another because of different scales adopted in convolution,resulting in multi-threshold segmentations of the image.Finally,a refined segmentation is gradually generated through the reiterative top-down logical operation.The method is implemented in 2 categories of ferrographic image,and the segmentation process shows that the top-down segmentation not only gets rid of uninterested regions,but also regains objective details on the smaller scale levels.Moreover,compared to the traditional thresholding segmentation,the proposed method has an increase by 14 percentage points in accuracy,and a decrease by 15 percentage points in false alarm rates.

关 键 词：铁谱图像 自适应阈值法 高斯尺度空间 自顶而下分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]