Level Set Image Segmentation Method Based On Shape Prior Knowledge

Compared with the traditional segmentation algorithm, the Active contour model shows great advantages which make itself a very popular and widely used segmentation algorithm in present image segmentation. However, this model still has defects, for example its obvious sensitivity to the initial contour and image noise active contour model due to the border-based feature that uses local information of gradient information near the border. Region-based active contour model takes advantage of general information and the initial contour can be arbitrarily placed, whereas it also shows defects, including that the result is poor when the image foreground and background intensity is similar or the intensity of target region is inconsistent. So a feasible research direction is concluded from the prior shape information into active contour model, while it is hard to meet in practical applications that the traditional segmentation method based on shape prior requires that the prior shape information must be the same as the target outline in position, scale and angle.An image segmentation method based on shape prior, is proposed to approach the defects described above. The main work and innovations are as follows:1. The combination of boundary and regional information enables the region and boundary features to affect the curve evolution at the same time; the introduction of prior shape information would guide the evolution of the profile curve to the target contour, and a new curve evolution model is constructed.2. Considering the location and scale of shape templates, during templates matching, we construct a novel similarity measure to guarantee the segmentation of translation invariance and scale invariance. It also can get ideal segmentation for partially obscuring images. Furthermore, the matched templates are also used in the generation of initial contour to facilitate the convergence of contour curve towards the target region. Compared with traditional way, the experiment proves that this approach contributes better segmenting effect and improved accuracy.3. The design and development of an algorithm evaluation system can be used for implementing and comparing the segmentation results of different algorithms and exert the analysis and comparison of algorithms by different evaluation criteria. To experiment with3data sets:the Weizmann set, a walking series of Atlas of Human and abdominal CT images, after the persuasive comparison and judgment of the new and traditional models, the experimental results confirm the validity of the new model.