A multiresolution flow-based multiphase image segmentation

In this work a variational model is proposed for simultaneous smoothing and multiphase image segmentation. By assuming that the pixel intensities are independent samples from a mixture of Gaussians, and by interpreting the phase fields as probabilities of pixels belonging to a certain phase, the model formulation is obtained by maximizing the mutual information between image features and phase fields. The proposed energy functional J(epsilon) consists of three parts: the smoothing term for the reconstructed image, the regularization for the boundaries in hard segmentation, and a likelihood estimator based on the density function. The segmentation and image denoising are performed simultaneously through the flow equation obtained by minimizing the energy functional with respect to the mixture of Gaussian coefficients and variance. Some experimental results on segmenting synthetic and natural color images are presented to illustrate the effectiveness of the proposed model.