Study On Region-based Active Contour Models For Image Segmentation

Image segmentation is a fundamental and important subject in image analysis and computer visions. The goal of image segmentation is to partition the image domain into a number of (mutually exclusive) subdomains, over which certain properties of the image appear to be homogeneous.Active contour models based on partial differential equations (PDEs) have been extensively applied to image segmentation. There are some desirable advantages of active contour models over classical image segmentation methods, such as edge detection, thresholding, and region grow. Existing active contour models can be categorized into two major classes: edge-based models and region-based models.In this dissertation, we first review a relative knowledge of mathematics and image segmentation based on PDEs. And then, we make respectively an improvement of C-V model and LBF model, which are two popular region-based active contour models.The main results of this dissertation are summarized as follows:1) C-V model based on gray level transformations. Power-law transformation and negative transformation are two basic types of gray level transformations used frequently for image enhancement. The proposed model incorporates the two transformations and C-V model to raise C-V model's performance on image segmentation. By practical experiments, it is verified that our model exhibits certain capability of handling intensity inhomogeneity and has faster convergence and more stable behavior than C-V model.2) Improvement of LBF active contours model. In contrast to C-V models, LBF model introduces a local binary fitting (LBF) energy with a Gaussian kernel function. Because the LBF energy enables the extraction of accurate local image information, LBF model can address the segmentation of images with intensity inhomogeneity, to which C-V models are not applicable. The proposed model utilizes a new kernel function instead of Gaussian kernel function. Experimental results show that the new LBF model is about 50% faster than the original LBF model.