基于阈值分割的多晶硅晶畴检测方法  

A Method for Domain Detection of Polysilicon Based on Threshold Segmentation

作　　者：黄世涛[1] 阎志军[1] 王丹[1] 

机构地区：[1]南京航空航天大学应用物理系,南京211106

出　　处：《现代计算机》2011年第24期28-32,共5页Modern Computer

基　　金：南京航空航天大学基本科研业务费专项项目资助(No.NS2010206)

摘　　要：0引言晶畴的区域分割提取是利用数字图像实现多晶硅结晶度检测中的一项关键技术,其中阈值的自动选择是实现自动化检测结晶度的核心。针对多晶硅图像的特点,充分利用图像的直方图信息,通过结合双峰法与波形对称性分析首先得到阈值数以及阈值的初步取值,然后利用最大类间方差法精确确定阈值,再进行硅片图像分割。实验证明,这种方法结合了双峰法和最大类间方差法的优点,在对多晶硅图像进行晶畴分割提取时能够获得优良的检测效果并明显提高算法的效率,为实现工业在线自动检测奠定坚实的基础。Digital image based domain segmentation is one of the key techniques for crystallinity detection of polysilicon. Automatic determination of thresholds is the core task to realize automatic machine detection. Based on histogram information and the characteristics of image, uses bimodal method and wave symmetry analysis to get the number of thresholds and their probable values, then utilizes Otsu＂ method to obtain the precise values of the thresholds. The experiment shows that this method combines the advantages of bimodal method and Otsu＂ method. Segmenting polycrystalline image with this algorithm can get well detection results and time consumption of this algorithm and has reduced to 1/3 - 1/4 compare with traditional. This method provides a solid foundation for the industrial automatic-line detection.

关 键 词：图像分割 晶畴区域 双峰法 峰谷信息 最大类间方差 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]