摘要
针对传统的图像显著性检测造成显著目标误检的问题,本文通过采用Harris角点检测算子以及计算机形态学中的凸包理论,得到改进的中心先验图像显著性检测系数,进而利用小波变换(WT)在时域和频域上的局部特征信息表征能力得到局部特征显著图像,以及采用谱残差(SR)的图像处理方法获取全局特征的显著图像,提出一种基于改进图像中心先验理论的自底向上的图像显著目标检测方法。实验结果表明,与现有的图像显著性检测模型相比,本文提出的模型检测准确率更好,检测效果也更好。
Aiming at the problem that the traditional image saliency target detection can cause false detection for saliency target,in this paper we find the center of target by Harris corner detection and convexhull of computer morphology theory, and built the advanced center-prior saliency detection model. Then, we get the local feature saliency image by the wavelet transform in the time domain and frequency do- main on the local characteristic information representation ability, and obtain the global features' saliency image by spectral residual to deal with target images. Finally, we establish an advanced center-prior saliency detection model which is based on bottom-up image saliency target detection methods. The experi- mental results show that the model proposed in this paper is effective in detecting significant object,the target detection rate is higher,and the detection of significant target is more complete.
出处
《光电子.激光》
EI
CAS
CSCD
北大核心
2016年第8期886-892,共7页
Journal of Optoelectronics·Laser
基金
国家自然科学基金(61203261
61273277)
山东省自然科学基金(ZR2012FQ003)
浙江大学CAD&CG国家重点实验室开放课题(A1514)
南京理工大学高维信息智能感知与系统教育部重点实验室创新基金(201501)资助项目