Research On WMSN Image Denoising Based On Sparse Regularization

Due to the unique advantage of rapid deployment, flexible networking and perceivingmultimedia information, WMSN technology has been widely used in many fields. However,the complexity environment, such as adverse weather or low light often leads the result thatWMSN images are corrupted by much noise and fail to meet the quality requirements.Therefore, image denoising becomes the key to ensure the reliability of the WMSNmonitoring.Image denoising is usually an ill-posed problem, and one popular approach to this problem isthe regularization method, which can utilize the prior of clean natural images to structure theregularization term of image denoising model, Turning the ill-posed problem into a well-posedproblem. Considering that sparse representation can fully describe the essential characteristics ofimage and can distinguish the original image from the noise according to whether the data can berepresented by atoms in the dictionary or not. So applying the sparse prior into image denoising modelcan effectively filter out the strong noise and retain more useful information. In this paper, research onWMSN image denoising based on sparse regularization plays an important role in ensuring thereliability of WMSN monitoring.Through further analysis of the WMSN image characteristics, and based on the methodof K-SVD image denoising, A image denoising via sparse regularization based on SSIM wasfirstly put forward. Unlike the former, the SSIM replacing the role of mean square error isemployed as the information fidelity of the sparity regularized denoising model. Experimentalresults showed that the proposed algorithm has greater ability to retain image structureinformation while removing much noise, and achieves better visual quality as compared toK-SVD.However, the aforementioned denoising methods only consider the sparse representationof a single image block, but fails to make full use of rhe possible structural similarity betweenimage blocks. This motivated us to propose a new sparity regularized denoising method basedon gradient histogram and non-loca self-similarity. The method makes full use of the sparsityand nonlocal NSS priors to construct the denoising model, and in the process of solving themodel, we also use the sparse K-SVD instead of PCA to run dictionary learning. Experimentalresults showed that the method can achieve better visual quality while removing much noiseand further reduce the computational complexity, suitable for the WMSN video imagedenoising.