Active Contour Models And Application To Biomedical Image

Image Segmentation plays very important role in image processing and computer vision, deciding the results of the following processing, for example, recognition and registration. However, the segmentation problems are yet unsolved, especially for medical image segmentation. Traditional segmentation algorithms face various difficulties when applied to medical images, like CT and MRI due to their nature. Active contour model/snake is one of the most successful segmentation methods in the field of medical and natural image segmentation. The main reasons can be explained as that the results of the active contour model are closed curves as well as their energy functions are flexible and their numerical solutions are very stable and easy-to-use.In this paper, we focus our research on the following aspects:1. We propose a novel active contour model to solve the problem of local minimum that the traditional active contour models suffer. The proposed model combines with the image bias field estimation to remove the image inhomogeneities. Firstly, we compute the bias field utilizing the Gauss function, which can ensure the bias field to keep smooth. Also, the piece constants inside and outside the curves can be calculated by exacting the local image information of the TV-norm, we can find the global minimum of the proposed energy rather than the local one. Experiments show that our model can obtain desire results both in nature and medical images. In addition, our model can segment the specific objects and cost much less computation than the traditional local-region model.2. The main drawback of the traditional level-set-based coupled curves evolution model is that it costs much computation and is very sensitive to noise. We propose a method using graph cut to minimize the normalized energy of the coupled curves evolution model which is based on the idea of maximizing the separation of the mean intensities of the two regions inside and outside the active curves. Using the Minimum cut method, experiments demonstrate the computation time of the proposed method is much less than the level-set-based method.3. Active contour is one of the most successful variational models in image segmentation, pattern analysis, and computer vision. However, traditional active contour models not only cost much expensive computation but are very sensitive to noise. In this paper, we propose a scheme for noisy image segmentation integrating the active contour model with the contourlet transform which is an optimal sparse representation of image. Having dandled the contourlet coefficients and reconstructed all the scale maps, we downsample the last but one scale map twice. Then we apply the active contour model on the coarsest scale map and take the segmentation results as the initial curves for the finer scale map. Experiments have demonstrated that our proposed method can make a desired segmentation results both in real images and synthetic images.