铁轨扣件图像自适应分割算法研究  

Research on Adaptive Segmentation Algorithm for Rail Fastener Image

作　　者：林晓函 何朗[1] LIN Xiao-han;HE Lang(School of Science,Wuhan University of Technology,Wuhan 430070,China)

机构地区：[1]武汉理工大学理学院,武汉430070

出　　处：《武汉理工大学学报》2023年第5期140-148,共9页Journal of Wuhan University of Technology

摘　　要：针对传统图像分割算法在铁轨扣件图像分割问题上的适应度低的问题,改进传统GrabCut算法,提出了一种图像自适应分割算法OTSUCut。首先,使用均值滤波消除图像噪声,避免图像噪声对图像分割结果的影响;其次,基于流形排序算法生成图像前后背景的先验信息,提高OTSU算法在分割前景时的分割结果;再次,构造图像的掩膜信息,生成目标图像的位置信息,代替GrabCut算法中的矩形框,实现图像的自适应分割;最后,基于前后背景的先验信息构造能量函数区域项的权重,优化迭代分割的过程。将OTSUCut算法与其它3种主流的图像分割算法在铁轨扣件数据集上进行实验对比,结果表明:在运行时间上,OTSUCut算法要快于其他算法,在分割后图像的直观对比上,OTSUCut算法在细节处分割效果比于其它算法更好,并且在SSIM、PSNR等一些评估指标上,OTSUCut算法分割结果相比于其它算法具有显著性的提升。Aiming at the low adaptability of traditional image segmentation algorithms in image segmentation of rail fasteners,an adaptive image segmentation algorithm OTSUCut is proposed by improving the traditional GrabCut algorithm.Firstly,using mean filtering to eliminate image noise and avoid the impact of image noise on image segmentation results;Secondly,apriori information of the background before and after the image is generated based on the manifold ranking algorithm to improve the segmentation results of the OTSU algorithm when segmenting the foreground;Thirdly,construct the mask information of the image,generate the position information of the target image,and replace the rectangular box in the GrabCut algorithm to achieve adaptive image segmentation;Finally,the weights of the energy function region terms are constructed based on prior information from the background to optimize the iterative segmentation process.An experimental comparison of the OTSUCut algorithm with three other mainstream image segmentation algorithms on a rail fastener dataset shows that the OTSUCut algorithm is faster in runtime than other algorithms.In terms of visual comparison of segmented images,the OTSUCut algorithm performs better in detail segmentation than other algorithms,and in some evaluation indicators such as SSIM and PSNR,the OTSUCut algorithm has a significant improvement in segmentation results compared to other algorithms.

关 键 词：铁轨扣件 图像分割 GRABCUT OTSU 流形排序 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]