Research On WMSN Video Denoising Methods Based On Dictionary Learning And Low-rank Matrix Reconstruction

As new wireless sensor devices with the integration of multimedia capabilities, WMSN can perceive, retrieve, handle and transmit multimedia data, which has caused wide attention of all sectors of society and shown its potential in many areas, such as environmental monitoring, traffic control, smart home and health care. With the rapid development of multimedia application, the quality demand of WMSN video increases. However, due to the complexity of WMSN monitoring scenes and inclement weather, the video images inevitably suffer from various noise types. Therefore, it is imperative to study denoising methods for removing mixed noise in WMSN video to ensure the reliability and robustness of WMSN video surveillance.For denoising problems, through in-depth analysis of the characteristics in WMSN video or images and the theory of low-rank construction, a mixed noise removal algorithm based on low-rank matrix completion with a fast block matching was designed. Firstly, some pretreatment is done on WMSN video by adaptive median filter in order to create favorable conditions for subsequent block matching. Consequently, an improved randomized correspondence algorithm was put forward to realize fast and accurate block matching and a low-rank matrix completion model was built after the similar matches were vectorized and stacked into a matrix. Finally, the appropriate optimization algorithm was choosed to solve the model and a denoised video will be reconstructed. Experimental simulation results show that the proposed algorithm owes the superiority compared with the classic VBM3 D method in every respect and gets good visual effect.However, the over-smooth phenomenon of flat area may cause the loss of some key details information after low-rank denoising model. On the strength of priori knowledge in image, a mixed noise removal method based on sparse K-SVD dictionary learning and weighted low-rank model was presented to deal with the aforementioned problem. Firstly, the distribution of mixed noise was fitted to approach gaussian distribution by adding a weighted value, which can be formed into a weighted low-rank denoising model solved by alternating minimization method. Then, take the sparse prior and nonlocal self-similarity prior into consideration, a nonlocal sparse representation denoising model was proposed, the denoised WMSN can be obtain by using sparse K-SVD dictionary learning algorithm. Experimental simulation results show that the proposed algorithm can preserve the detail information in WMSN video effectively and gets a better visual performance.