基于K-SVD的低信噪比WMSN视频图像稀疏去噪  

K-SVD based sparse denoising for WMSN video image with low SNR

作　　者：罗晖[1] 褚红亮[1] 王世昌[1] 

机构地区：[1]华东交通大学信息工程学院,江西南昌330013

出　　处：《计算机工程与科学》2014年第3期497-501,共5页Computer Engineering & Science

基　　金：国家自然科学基金资助项目(61261040)

摘　　要：无线多媒体传感器网络WMSN因感知视频等信息的优势而被广泛应用,但受天气、光照等外因干扰,所采集视频图像常含有较为严重的噪声。因此,在低信噪比条件下进行视频图像去噪是保证WMSN视频监测有效性和可靠性的关键。在分析WMSN视频图像特征的基础上,首先对其进行周期性采集、分帧及帧差等预处理;然后对关键帧运用K-SVD训练DCT冗余字典以充分稀疏表示图像特征,并采用基于残差比的改进型Batch-OMP实现关键帧去噪及重构,而对残差帧则基于DCT冗余字典进行稀疏去噪处理;最后,叠加去噪后的关键帧和残差帧,从而整体上实现低信噪比WMSN视频图像的去噪及重构。实验表明,本算法能更加有效地、较为快速地滤除视频图像噪声,适用于低信噪比WMSN视频图像去噪。As a highly effective method of perceiving multimedia information, Wireless Multimedia Sensor Networks （WMSNs） has shown its potential in many areas. However, the outside interference in the monitoring environment brings severe noise to video images. Obviously, video image denoising be- comes the key to ensuring the validity and reliability of WMSN video monitoring. To denoise WMSN video image, firstly, its features are analyzed and some pretreatment are done. Secondly, the K-SVD al- gorithm is employed to adaptively train DCT dictionary for reflecting the image characteristics and recon- struct the key frame through improved Batch-OMP algorithm with residual ratio as the iteration termina- tion, while DCT dictionary is adopted to sparsely denoise the residual frames. Finally, the video image is reconstructed under the situation of low SNR. Experimental results show that, compared with its coun- terparts, the superiorities of the algorithm can be observed in both visual and some numerical guidelines, showing the suitability for the WMSN video image denoising in low SNR.

关 键 词：稀疏去噪 K奇异值分解 残差比 低信噪比 无线多媒体传感器网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]