A Study On Total Variational Image Denoising Algorithms

Image denoising is the basic problem of image processing and a hot issue in the field of signal processing.Only by removing the noise in the image as much as possible can the image clarity and information accuracy be satisfied,and the higher level processing and analysis of the image can be performed.With the development of denoising technology,the total variation denoising method has become an important method for image denoising with its powerful mathematical support and excellent image edge preservation.Most of the existing total variation denoising models only consider the image gradients in the horizontal and vertical directions,and there are step effects in the smooth region.The existing algorithms have the problem of too many iterations and too long time.In this paper,the existing total variation model and numerical solution method are analyzed,and we proposed a quaternion relative total variation model,which improves the denoising effect,solves the step effect,and gives a fast solution.The iterative weighted least squares method improves the efficiency of the algorithm.This paper mainly completed the following work:● In order to solve the step effect,a Gaussian weighted operator is introduced to smooth the image gradient,which reduces the gradient mutation of the smooth region in the result graph.In terms of the image edge retention,this paper uses the inherent variable partitioning noise and image structure information to remove the noise while preserving the image details as much as possible.● In addition to the image gradients in both horizontal and vertical directions,the gradients of 45° and 135° are also added to the regular term of the model,making the denoised image more in line with the visual requirements of the human eye.● The iterative weighted least squares method is proposed to solve the model,which accelerates the convergence speed.● From the two aspects of the optimal solution and computation platform of the linear equation system,the process of solving the total variation model by the iterative weighted least squares method is optimized and accelerated,and the GPU is used to accelerate the calculation process,which reduces the consumption time of the algorithm and satisfies efficiency requirements in practical industrial inspection applications.The quaternion relative total variation model proposed in this paper has achieved good denoising effect in both standard dataset and industrial atlas.It is verified a general denoising algorithm,and the time-consuming of the algorithm meets the practical application requirements.