Performance Comparison of Total Variation based Image Regularization Algorithms

The mathematical approach calculus of variation is commonly used to find an unknown function that minimizes or maximizes the functional. Retrieving the original image from the degraded one, such problems are called inverse problems. The most basic example for inverse problem is image denoising. Variational methods are formulated as optimization problems and provides a good solution to image denoising. Three such variational methods Tikhonov model, ROF model and Total Variation-L1 model for image denoising are studied and implemented. Performance of these variational algorithms are analyzed for different values of regularization parameter. It is found that small value of regularization parameter causes better noise removal whereas large value of regularization parameter preserves well sharp edges. The Euler’s Lagrangian equation corresponding to an energy functional used in variational methods is solved using gradient descent method and the resulting partial differential equation is solved using Euler’s forward finite difference method. The quality metrics are computed and the results are compared in this paper.