视频压缩感知中组稀疏表示的自适应阈值算法

Adaptive threshold algorithm for group sparse representation of compressed video sensing

在对帧间组稀疏表示框架研究后,提出一种改进的组稀疏表示的自适应阈值算法(AT-GSR)。在变换域进行阈值处理过程中,根据采样率,在迭代开始时对初始阈值进行自适应设置,在迭代过程中对阈值进行阶梯型递减,保证信号在噪声被滤去的前提下,保留更多细节特征。针对非剧烈运动序列提出使用重构精度较高的关键帧作为参考帧的方案,保证帧间匹配块的精度,且利用前后两个方向的时间相关性。仿真结果表明,所提重构算法AT-GSR,对于运动不太剧烈的视频序列,相对于SSIM-InterF-GSR降低了算法复杂度,提高了重构性能,与目前性能好的其它两种视频压缩感知算法相比,性能也有明显提升。

After researching on interframe group sparse representation framework,an improved group sparse representation adaptive threshold algorithm(AT-GSR)was proposed.In the process of threshold processing in transform domain,the initial threshold was set adaptively according to sampling rate at the beginning of iteration,and the threshold was stepped down during iteration,which guaranteed that the signals retained more details while the noise was filtered out.For non-violent motion sequences,a scheme using key frames with high reconstruction accuracy as reference frames was proposed,which not only guaranteed the accuracy of inter-frame matching blocks,but also utilized the temporal correlation of the forward and backward directions.The simulation results show that the proposed reconstruction algorithm AT-GSR improves the reconstruction performance and reduces the algorithm complexity compared with SSIM-InterF-GSR.Compared with the other two outstanding compressed video sensing algorithms,the performance of AT-GSR is also significantly improved.