Image Segmentation and Selective Smoothing Based on Variational Framework

This paper addresses the segmentation and smoothing problems in biomedical imaging under variational framework. In order to get better results, this paper proposes a new segmentation and selective smoothing algorithm. This paper has the following three contributions. First, a new statistical active contour model (SACM) is introduced for noisy image segmentation. SACM is proposed to solve the problem in fast edge integration (FEI) method, which takes advantages of both edge-based and region-based active contour model but only considers the mean information inside and outside of the evolution curve. In SACM, a new statistical term for considering the probability distribution density of regions and a unified variational framework are proposed for construction of different segmentation models with different probability density functions. Moreover, a penalized term is also introduced in the proposed model as internal energy in order to avoid the time consuming re-initialization process. Second, a new symmetric fourth-order PDE denoising algorithm is developed to avoid the blocky effects in second-order PDE model, while preserving edges. Third, in each stage of segmentation process, different denoising algorithms (or different parameters in the same denoising model) can be employed for different sub-regions independently, so that better segmentation and smoothing results can be obtained. Compared with existing methods, our method is more flexible, robust to noise, computationally efficient and produces better results.