摘要
传统的分布式视频压缩感知(distributed video compressive sensing,DVCS)系统有效降低了无线视觉传感器网络(wireless visual sensor network,WVSN)的传输成本和编码器复杂度,但在涉及快速运动或复杂纹理的动态场景中表现不佳.它们对时域相关性和跳块的依赖导致重建伪影和结构细节丢失,尤其是在高跳块率下更加明显.本文提出采用运动-纹理联合(motion-texture joint,MTJ)阈值跳块算法来提升DVCS 性能,其中运动模块先通过绝对差值和(sum of absolute differences,SAD)对候选跳块进行粗筛,再结合帧插值与运动矢量生成稳健性更好的边信息,用于识别冗余内容并辅助跳块决策.此外,基于局部二值模式(local binary pattern,LBP)分析的纹理特征模块,利用相邻帧的 LBP 纹理特征、方差纹理和梯度能量纹理,提升帧间相关性.在此基础上,运动-纹理联合自适应权重模块根据其时空贡献程度来动态分配运动信息和纹理信息的权重系数,并控制 DVCS 的跳块率.在标准视频序列上的实验表明,所提 MTJ算法相较最先进的统一参考阈值动态跳块(uniform-reference threshold-dynamic skipping,UTS)算法,平均峰值信噪比(peak signal-to-noise ratio,PSNR)提升 0.12 dB,编码器复杂度仅微增.尤其在 70%跳块率下,该算法在 Foreman 序列上相对 UTS 的 PSNR 最大提升达到 0.20 dB.MTJ 算法有效平衡了重建质量和计算复杂度,使其更适用于 WVSN.
The traditional distributed video compressive sensing(DVCS)system effectively reduces transmission costs and encoder complexity for wireless visual sensor networks(WVSN),but struggles with dynamic scenes involving rapid motion or complex textures.Their reliance on temporal correlation and skipping causes reconstruction artifacts and loss of structural details,especially at high skip ratios.This paper proposes the motion-texture joint(MTJ)algorithm to enhance DVCS performance,where the motion module first preliminarily screens those candidate skip-blocks using the sum of absolute differences(SAD)and then generates more robust side information by combining frame interpolation with motion vectors for redundant-content identification and skip-block decision.Further,a texture feature module based on local binary pattern(LBP)analysis jointly exploits LBP texture features,variance textures,and gradient energy textures from adjacent frames,which improves inter-frame correlations.On this basis,an MTJ adaptive weight module dynamically allocates coefficients to motion information and texture information according to their spatio-temporal contributions and controls a skip ratio of DVCS.Experiments on standard video sequences demonstrate that the MTJ algorithm achieves an improvement in average peak signal-to-noise ratio(PSNR)of 0.12 dB over state-of-the-art uniform-reference threshold-dynamic skipping(UTS)with minimal encoder complexity increase.Especially at a 70%skip ratio,the maximum PSNR improvement over UTS reaches up to 0.20 dB on the Foreman sequence.MTJ algorithm effectively balances reconstruction quality and computational complexity,making it suitable for WVSN applications.