A Multi-frame Image Speckle Denoising Method Based on Compressed Sensing Using Tensor Model

Due to the bad channel environment and poor image sampling equipment, images are often contaminated by noise in the process of collection, transmission and processing. Speckle noise, which is difficult and complex to eliminate, is one of the common noise appearing in image processing. Denoising methods based on Compressed Sensing (CS) technology have been proved as useful tools in suppressing speckle noise of single-frame images. However, temporal correlation in multi-frame images has not yet been utilized. Considering that the traditional denoising methods do not work satisfactorily in speckle noise reduction, a multi-frame image speckle denoising methods based on compressed sensing using tensor model is proposed. The first step is to use the third-order tensor to represent the blocks of image sequences, then the denoising tensor model is established according to the CS theory and the corresponding optimization problem is raised. The problem is divided into three parts: the sparse representation, the tensor dictionary update and the image reconstruction. A Kruskal tensor-based Orthogonal Matching Pursuit (OMP) and Candecomp/Parafac (CP) analysis are used to solve these problems and get the denoised image. At last, simulations are conducted to compare the CS method and traditional methods. It is shown that the CS-based multi-frame speckle denoising method performs well in noise variance and can significantly enhance the visual quality of the image.