摘要
针对现有的感兴趣区域(ROI)提取方法边缘不清晰、区域不完整等问题,提出一种ROI提取方法.首先采用颜色局部特征的信息量大小衡量兴趣度的大小,然后融合颜色特征信息量图获得图像的显著图(SM),再进行阈值分割,得到显著值大的区域,即ROI.实验结果表明,该方法能有效地提取出感兴趣的对象,SM中目标区域的显著值均匀、边缘清晰;与人工标记的ROI比较,该方法召回率为79.71%,精度为78.53%,优于已有的ROI提取方法.
Ill-defined object boundaries and imperfect extracted objects are the major problems existing in interest region detection methods. To address those problems, a new approach for interest region detection based on the amount of color feature information calculation is proposed. Through calculating the amount of color feature information, three color information maps can be obtained firstly. Then a saliency map will be generated by combining three information maps linearly. After thresholding and denoising, interest regions are highlighted. Experiment shows that the proposed method can highlight interest regions perfectly with well-defined boundaries. Additionally, compared with human labeled ground truth, the proposed method reaches 79.71~ in recall and 78.53G in precision, which is higher than the existed methods.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2013年第6期852-856,共5页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(60970098
61173122)
湖南省自然科学基金重点项目(12JJ2038)
湖南省自然科学基金(11JJ3067)
浙江大学CAD&CG国家重点实验室开放性课题基金(A1011)
关键词
感兴趣区域
显著图
颜色特征
特征融合
信息量
region of interest
saliency map
color feature
feature combination
the amount of information
