摘要
运动估计在视频编码中占据十分重要的地位,因为它能有效地去除图像序列的时域冗余,大大提高编码效率,但往往算法复杂,占用整个编码过程较多的时间。针对智能视频监控等领域的应用,提出一种基于视觉注意力的快速运动估计算法。该算法的核心思想是利用视觉注意力模型将关注的运动区域提取出来作为显著性区域,然后利用快速运动估计算法在显著性区域采用较高的预测精度,而非显著性区域采用较低的预测精度。实验结果表明在峰值信噪比几乎不变的情况下,该算法比其他经典运动估计算法的运算时间有明显减少。
Motion estimation plays a very important role in video coding because it can eliminate the temporal redundancy effectively to achieve good coding efficiency. However, the process of motion estimation costs a great majority of coding time for its complexity. In this paper, a kind of fast motion estimation algorithm based on visual attention is given for smart video monitoring. The essential idea of the algorithm is as follows： a simple model of visual attention is utilized to extract the motion attention regions as saliency regions. The prediction precision is high in saliency regions and is lower in non-saliency regions. Experimental results show that the algorithm proposed in this paper reduces the computation time significantly compared with other traditional methods while maintaining very similar PSNR.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2009年第12期2539-2543,共5页
Chinese Journal of Scientific Instrument
基金
安徽高校省级自然科学基金(KJ2009B035)
芜湖市科技计划重点项目(2008320)资助
关键词
视觉注意力
运动预测
视频编码
visual attention
motion estimation
video coding
