基于投影直方图提取目标感兴趣区域的新方法  

Region of Interest Extraction Algorithm Based on Gray-level Projection Histogram

作　　者：高敏[1] 吴福培[1,2] 李昇平[1,2] 

机构地区：[1]汕头大学工学院机电系,广东汕头515063 [2]汕头轻工装备研究院,广东汕头515063

出　　处：《汕头大学学报（自然科学版）》2012年第4期54-60,共7页Journal of Shantou University：Natural Science Edition

基　　金：国家自然科学基金(51175315);广东省重大科技专项(2010A080402010);广东省产学研引导项目(2011B090400474);汕头市科技计划项目(B201100456)

摘　　要：提出了一种基于灰度投影直方图获取单幅检测样品的目标感兴趣区域的方法,并利用鲁棒的图像匹配算法生成参考图像.首先,采用改进的灰度投影直方图法获得单幅图像的检测样品感兴趣区域,建立感兴趣区域提取模型;其次,利用基于形状的特征匹配算法学习样品图像,生成印刷品检测的参考图像,在此基础上,提出基于图像金字塔数据结构的分层匹配方法.实验结果表明,该方法能够在87 ms内获取首张图像样品感兴趣区域,并且图像匹配过程对图像之间存在的明暗程度、旋转变化等干扰具有良好的鲁棒性,匹配精度可达95%,匹配时间为152 ms,可快速准确地生成参考图像.A method for achieving region of interest（ROI） of single detection sample and generating reference image is proposed, which based on gray-level projection histogram and robust image matching algorithm. Firstly, the single detection sample ROI is obtained by improved gray-level projection histogram, and the model of extracting ROI is built. Secondly, the reference image is built by studying sample images, which are trained by the image feature matching algorithm. A layered matching method is proposed based on pyramid data structure. Experimental results demonstrate that the proposed method can acquire the ROI of the first image sample within 87ms, and it shows the good robustness to uncertain disturbance, which originates from the rotation and noise of image. Furthermore, the matching accuracy and time are 95%, 152ms, respectively. The results prove that the proposed method can generate reference images accurately and quickly.

关 键 词：检测样品 感兴趣区域 灰度投影直方图 图像匹配 参考图像 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]