摘要
为解决当前图像目标检测算法中存在的虚警率和漏检率较高的问题,提出基于特征的多媒体网络图像目标检测算法。分别计算目标辐射信号和白噪声相互叠加产生的信号和小目标像素强度,得到小目标中噪声边缘点像素强度。将噪声边缘点像素强度引入最大均值滤波中,给出最大均值滤波器在不同情况下的输出和多媒体网络图像最终滤波结果。将滤波后多媒体网络图像划分为相等的图像块,计算各图像块像素点总数,得到目标弱可疑区域。在弱可疑区域实行非完全特征匹配操作,获得强可疑区域,实行完全特征匹配操作,计算出多媒体网络图像目标质心位置。利用目标质心位置的确定,给出多媒体网络图像目标检测结果。仿真结果表明,所提算法可将漏检率控制在1. 5%以下,检测虚警率最高不超过5%。实验数据说明,所提算法具有很强的可实践性。
In order to solve the high false alarm rate and missed rate in the current image target detection algorithm,this paper puts forward an algorithm to detect targets in multimedia network image based on features.Firstly, this research respectively calculated signals caused by superposition of target radiation signal and white noise and the pixel intensity of small target,and then the pixel intensity of noise edge in the small target was obtained.Secondly, our research introduced the pixel intensity of noise edge point into the maximum mean filter,and then.outputs ,of the maximum mean filter in different conditions and the final filtering result of multimedia network image were given.In addition,the research divided the filtered multimedia network image into equal image blocks,and then calculated total of pixel points in each image block to obtain the weak suspicion area.In the weak suspicion area,we carried out the incomplete feature matching operation and thus to obtain strong suspicion area.Meanwhile,we carried out the complete feature matching operation and thus to calculate the objective centroid position of multimedia network image.Finally,the determination of objective centroid position was used to obtain the target detection result of multimedia network image.Simulation results prove that the proposed algorithm can keep the missed rate below 1.5%. Meanwhile,the maximum false alarm rate is less than 5%.Simulation data shows that the proposed algorithm has strong practicality.
作者
刘蓉
李红艳
LIU Rong;LI Hong-yan(Changsha Medical University,Changsha Hunan 410219,China)
出处
《计算机仿真》
北大核心
2018年第12期346-349,405,共5页
Computer Simulation
基金
湖南省教育厅优秀青年项目(15B030)
关键词
多媒体网络图像
目标
检测
Multimedia network image
Target
Detection
