The Chan-Vese Segmentation Algorithm and Global Properties of an Image

643 Words2 Pages

The Chan-Vese segmentation algorithm is robust and has been used to segment different kind of images. This algorithm re- lies on global properties of an image (gray level intensities in regions, length of contours, area of regions), hence it is more suitable in cases when the edge information is not very predominant. The results are good qualitatively for noisy im- ages and images with complicated topologies. Reviewing the state of the art techniques for segmentation we see that this method is used extensively in segmentation of regions in med- ical images. Images with low-contrast do not have well de- fined homogeneous regions, performance may be degraded in these scenarios since the method assumes homogeneous in- tensities in the foreground and background. Level-sets being non-parametric, include a regularization term such as penalty on curve/length/surface area or curvature. These regulariza- tion terms do not contain any information about the shape of the region of interest. The Chan-Vese algorithm may be slow for some applications. Depending on the type and size of the image and the number of iterations needed, the segmentation can be slow, hence, GPU implementations of the algorithm, reduction of the segmentation problem to an ordinary differ- ential equation rather than a nonlinear PDE [13] can be used to speed up the algorithm. in 3D. Presently the model contour segment parts are selected and grouped into bundles manually. It is important to auto- mate the step to learn part bundles. Timing performance of the algorithm is not discussed in the paper.
4. FUTURE WORK
4.1. Active Contours Without Edges
4.1.1. Parameter Selection and Initialization of the Level Set
Function
This method is semi-automated. It requires user interaction for tuning the parameters and initializing the level set. To ob- tain qualitatively good segmentations the parameters need to be tuned for every image. Machine learning algorithms can be used to learn parameter values for homogeneous images from a training set of images. Unfortunately, real images are not homogeneous, there may be variations in the image due to varying lighting conditions. Methods devised to tune these parameters automatically would benefit the segmentation pro- cess. Careful initialization of the level set is necessary, since it has a direct correlation with the time taken for convergence.
Atlas-based initializations can also be used to segment struc- tures from medical images.
4.1.2. Improve efficiency
Accuracy and precision can be improved by incorporating prior information about the target to be segmented. For in- creasing computational efficiency multiscale processing and parallelizations are viable solutions.
3.2. Shape Guided Contour Grouping with Particle Fil- ters The contour grouping algorithm was applied to the ETHZ shape classes. The ETHZ shape classes have significant intra- class variations, scale changes, and illumination changes. Re- sults were shown in the presence of a lot of clutter in the

More about The Chan-Vese Segmentation Algorithm and Global Properties of an Image

Open Document