Variational image processing has been studied extensively thanks to its strong mathematical theory and existence of state-of-the-art numerical PDE methods. In this dissertation, we present a piecewise constant image segmentation model based on a new implicit curve evolution technique in the context of variational approach. In our new approach, we use multiple level sets of an evolving level set function to represent boundaries among objects. We extend the piecewise constant Chan-Vese segmentation model by combining their model with multilayer level set approach. This new approach is applicable for images with known topology and nested structure. e.g. MR brain images. By construction, our approach requires fewer number of level set functions than the Chan-Vese approach does, resulting in lowering computational cost. We show how we can apply the multilayer segmentation model to 3D MR brain data. Moreover, we present ways of incorporating brain atlas into our variational multilayer image segmentation model and show numerical results as well as comparisons with existing segmentation algorithm. We also discuss different choices of regularization as ways to keep the level set function more regular. Finally, we propose a region based implicit active contour model for image denoising, segmentation and deblurring, and discuss how this model can be used to model images with intensity inhomogeneity and blurry boundaries, such as MR brain images.