摘要
针对时空显著性框架的融合问题,提出了一种基于剪切波融合的时空显著性检测算法。先获取视频帧的空间和时间显著图,再分别对空间和时间显著图进行剪切波分解,获取系数。采用一定的机制融合对应的剪切波系数和尺度系数,通过剪切波逆变换,得到综合显著图,实现了视频的时空显著性检测。结果表明,该算法能够较好地利用空间和时间显著图提供的信息,对显著对象内部区域的标注能力更强,同时对空间和时间显著图携带的噪声具有更好的鲁棒性。
To tackle the fusion issue of spatio-temporal saliency framework,a Spatiotemporal saliency detection model via shearlet based fusion mechanism was proposed.Spatial and temporal saliency for each video frame was firstly detected.Then,the spatial and temporal saliency maps were decomposed by shearlet transform to obtain coefficients.After that,by properly fusing shearlet and scalling coefficients respectively,the final saliency map was obtained via inverse shearlet transform,which is the spatiotemporal saliency detection result for input video.Experiments show that the proposed mechanism can properly use the information in spatial and temporal saliency maps,with better performance on detecting pixels inside salient objects,and is more robust to noise compared with other models.
出处
《解放军理工大学学报（自然科学版）》
EI
北大核心
2016年第1期8-16,共9页
Journal of PLA University of Science and Technology(Natural Science Edition)
基金
江苏省自然科学基金资助项目(BK2012512)
中国博士后科学基金资助项目(2013M532205)
关键词
时空显著性检测
剪切波变换
图像融合
多尺度分析
spatio-temporal saliency detection
shearlet transform
image fusion
multi-scale analysis
