A Robust Segmentation Algorithm for Images with Intensity Inhomogeneity

As for the inhomogenous images, it is difficult and ineffective to segment Regions Of Interest (ROI). In order to solve these problems, this paper proposes an image segmentation algorithm based on the active contour model. Different from the ones in traditional level set techniques, which only use single information, a new energy function is defined by combining object edge information and regional statistical information. Utilization of edge information is in favor of the contours evolving into the object boundaries quickly and accurately. Regional statistical information consists of both local and global statistical information inside and outside the evolving contours. On the one hand, utilization of local region information facilitates the method to deal with intensity inhomogeneity. On the other hand, using global region information can avoid the evolved contour trapping into the local minima. In addition, in the evolution process of the contour, a Gaussian filter is adopted to quickly regularize the level set function, which avoids an expensive computational re-initialization or regularization. Experimental results using synthetic and real images show that the proposed approach can not only effectively segment objects with the weak boundaries in inhomogenous images, but also accurately segment the complex structure objects with multi-gray levels. At the same time, the method is robust to noise and the initial contour.