面向语义的图像中主要对象的提取方法  

Extraction Approach of Principal Objects in Image

作　　者：孙超[1] 周海英[1] 

机构地区：[1]中北大学电子与计算机科学技术学院,山西太原030051

出　　处：《电视技术》2013年第5期45-48,共4页Video Engineering

摘　　要：图像中主要对象的提取对于图像语义的抽取以及图像内容的自动标注具有重要作用。提出了一种面向语义的图像主要目标的提取方法。对于分割后的图像,计算不同区域的显著相关色得到核心目标,然后从中提取训练样本集,通过人工神经网络对图像中的块进行分类训练,获得各个区域的主要对象。实验表明,该方法可以较准确地提取出中心区域的对象以及周围区域的对象,并能保留图像的主要信息。Extraction of image principal objects is of im^x）rtant role both to image semantic extraction and automatic image annotation. A semantic-ori- ented extraction approach of principal objects is proposed. In segmented image,by means of signifieantly correlated color computation in different regions the core objects are obtained and from which the training sample set is extracted. The image patches is trained and classified by artificial neural network, and then the principal objects in regions are extracted. The experiment shows that this approaeh can extract the object in central region and the surround- ing of objeets precisely and keep the information of them.

关 键 词：颜色相关直方图 显著相关色 神经网络 图像划分 

分 类 号：TN911.73[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]