Based on variational level set image segmentation algorithm
|School||Nanjing University of Technology and Engineering|
|Keywords||Image Segmentation Partial Differential Equations Variational level set Energy functional|
Image segmentation in pattern recognition and computer vision plays an increasingly important role, in which segmentation method based on partial differential equations with its unique advantages gained widespread attention. Such methods are mainly divided into two categories based on edge and region-based model. PDE edge-based segmentation method is to use the edge of the target image information, such as gradient, curvature, etc. to be split, but such method can not solve the problem of curve topology changes. Region-based segmentation method of partial differential equations that active contour model, the idea of ??using the level set, the use of higher-dimensional space curve evolution, can be a good deal curves split or merge. With the active contour model of continuous development and improvement, such model has been fully developed and perfected. In this paper, active contour models are appropriate improvements, resulting in better image segmentation results. First introduced the basics of image segmentation, including some mainstream segmentation method, and the corresponding mathematical basis; then focuses on the level set image segmentation algorithm and these classic model experiments were carried out, a brief analysis of the advantages and disadvantages of each method ; Finally, on the basis of existing algorithms from two aspects of the model is improved. This work and the results obtained are as follows: (a) presents both global optimal solution, and make full use of local image information variational level set image segmentation algorithm. Multi-contour CV model with the characteristics of the global optimum, while the LBF model in an iterative process through the introduction of Gaussian kernel image cleverly local information, this paper combines the advantages of these two models, proposed a multi-contour CV-LBF model. This model without re-initialization can take full advantage of local information, but also has the characteristics of the global optimum. Finally it is extended to the multi-phase situation. Through experimental verification of the proposed algorithm. (2) in a multi-level set silhouette single model based on improvements in its external energy terms, the model contains the curve of the edge detection factor. This model not only can be split more different target gradation, and the full use of the edge curve of the target edge information, the edge detection, so that more accurate segmentation results. Experimental results show that the model for the weakly split edges have a significant effect.