A Convex Relaxation Model Of Image Segmentation For Images With Noises

Image segmentation is an important research aspect in computer vision. Image segmentation for images with noises has always been concerned. As a matter of fact, images with noises consistent with different probability distribution model, e.g., in synthetic aperture radar(SAR), where the noise is assumed to follow a Gama distribution, K-distributed or Rician distributed noise, in ultrasound imaging, where we are confronted with Rayleigh distributed, in electronic microscopy, single particle emission computed tomography(SPECT) and positron emission tomography(PET), Poisson noise appears in connection with blur. On one hand, the traditional variational level set model is a local energy model and the segmentation result depends on the initialization, on the other hand, the complexity of difference scheme leads to huge calculation and low efficiency. So the improved segmentation models and rapid algorithms are designed in this paper. A general model of two phase segmentation of images with noises based on binary label function and convex relaxation method is proposed. Its region model is based on the general probability distribution functions including Gaussian, Rayleigh, Poisson and Gamma distribution, and so on. The length term of active contour model is approximated using the total variation of the binary function. In the alternate process of optimization the parameters are estimated. The global minimization is realized via convexification and thresholding techniques, and the Split Bregman algorithm is designed.