Abstract: To solve the problem caused by intensity inhomogeneity in medical images, this paper
proposed an active contour model based on the local polarity information. By incorporating the local
information of image, the proposed model can efficiently segment the image which intensity is
nonuniform. Through introducing a penalizing energy term into the regularization, the level set
function can be approximated as a signed distance function all the time in the process. This algorithm
regarded the image segmentation as the task to minimize the curve of energy functional. Firstly, to
establish the evolution curve of energy functional including the local gray level information (polarity
information) and the improved signed distance function. Then to solve the minimization of energy
function by using the variational level set method to be able to get the final segmentation result. The
experiment results of real medical images and artificial synthesis images have shown that our method
can get a higher precision of segmentation for the medical images with uneven grayscale. In addition,
the image processing speed has been improved greatly. Because of the utilization of local gray
information, we can segment the medical image with uneven grayscale effectively, the improved
variational level set can completely avoid the re-initialization, and greatly improve the efficiency of
image segmentation.

Key words: variational level set, image segmentation, symbolic distance function, polarity information