基于自适应阈值的齿轮干涉图像前景区域提取方法  

Extraction method of gear interference image foreground region based on adaptive threshold

作　　者：赵逸超 王晛[1] 焦明星[1] ZHAO Yichao;WANG Xian;JIAO Mingxing(School of Mechanical and Precision Instrument Engineering,Xi'an University of Technology,Xi'an 710048,China)

机构地区：[1]西安理工大学机械与精密仪器工程学院,陕西西安710048

出　　处：《应用光学》2023年第2期345-353,共9页Journal of Applied Optics

基　　金：中国博士后科学基金(2020M683683XB);陕西省自然科学基础研究计划(2022JQ-403);陕西省教育厅自然科学专项(20JK0810);西安市碑林区科技计划(GX2112)。

摘　　要：齿面物体像灰度法是激光移相干涉测量中提取齿轮干涉图像前景区域的重要方法之一,针对该方法因人工设定阈值且忽略不同图像边缘特征从而导致的测量效率及精度受限问题,提出了一种基于自适应阈值的齿轮干涉图像前景区域提取方法。首先分析齿轮齿面形貌特征与各边缘顶点差异,对图像进行区域划分;然后根据边缘灰度变化规律通过邻域窗口筛选合格像素点并获取掩模结果,实现前景区域提取;最后根据5类图像评价指标分别对4组算法分割结果与传统方法分割结果进行数据对比。结果表明:算法在实现图像自动处理的基础上与参考结果匹配精确度提升约3.5%~4.5%,PRI(probabilistic rand index)提升约3%~4%,VOI(variation of information)提高约15%~25%,GCE(global consistency error)降低约2.5%~3.5%,最终相位信息准确度提升9μm~15μm。结果符合精度要求,该方法可广泛应用于齿轮干涉图像前景提取中。Gray method of tooth surface objects is one of the important methods to extract the gear interference image foreground region in laser phase-shifting interferometry. In view of the problem of limited measurement efficiency and limited accuracy caused by the manual threshold setting and neglect of different image edge features, a foreground region extraction method of gear interference image based on adaptive threshold was proposed. Firstly, the morphological characteristics of gear tooth surface and difference of each edge vertex were analyzed, and the image was divided into regions. Then, according to the changing rule of edge gray scale, the mask results were obtained by selecting qualified pixels through neighborhood window, so as to realize the foreground region extraction. Finally, the segmentation results of four groups of algorithms and traditional methods were compared according to five kinds of image evaluation indexes. The results show that the matching accuracy between the algorithm and reference results is improved by about 3.5%~4.5% based on the automatic image processing, the probabilistic rand index(PRI) is improved by about 3%~ 4%, the variation of information(VOI) is improved by 15%~25%, the global consistency error(GCE) is reduced by2.5%~3.5%, and the final phase information accuracy is increased by 9 μm~15 μm. The results meet the requirements of accuracy, and the method can be widely used in foreground extraction of gear interference image.

关 键 词：齿轮齿面形貌 图像分割 图像灰度 图像评价 

分 类 号：TN29[电子电信—物理电子学]