Simultaneous multiphase image segmentation and cartoon-texture decomposition

Conventional multiphase image segmentation methods perform well in segmenting piecewise constant or local piecewise constant images. However, real images usually contain intensity variation induced by illumination inhomogeneity, small scale details, texture and noise. This brings great challenges to image segmentation and heavily degenerates the performance of the conventional methods. In this work, we proposed a new fuzzy segmentation model which unified image segmentation and cartoon-texture decomposition. We also presented an iteration algorithm for solving the proposed model. In the model, the decomposition and the segmentation tasks executed alternatively and affected each other. The cartoon component resulted from the decomposition is a piecewise smooth approximation of the original image and it only contains the main structure and large scale edges of the original image. Segmentation is essentially performed on the cartoon part, thus it can avoid the influence of intensity variation caused by small scale details, texture and noise. Extensive experiments confirmed that the proposed method can effectively overcome the drawback of conventional methods and performed much better than conventional methods in segmenting images with strong intensity variation caused by small scale details, texture and noise.