Segmentation Of Abnormity Target And It’s Application To Embedded System

Image segmentation is a sub-region of the computer vision and it is thefoundation of the image analyses and recognition. The result of image segmentationhas a direct affect on the image process afterward. Recently in the domain of imagesegmentation, active contour model based on energy minimization has been studiedbroadly. The main work of this paper contains geometrical active contour model alsocalled level set method, and the former level set method is improved. As we allknow, the embedded system is broadly implemented in many industrial products. Inthis paper, the improved segmentation method is applied in the embedded system webuild up for this paper.Firstly, several famous level set methods are discussed in this paper. And thelevel set methods are improved from two aspects. On one hand, the distanceregularized level set method proposed by Li et al. has the disadvantage of requiringthe initial curve around or inside the objects to be detected. And it can not deal withintensity heterogeneity in the process of segmentation. In this paper, an adaptivedistance regularized level set method is presented, in which the initial curve can startanywhere in the image, and without the need to place the initial curve completelyexterior or interior to the real object boundaries. The adaptive distance regularizedlevel set method allows for more flexible applications. The improved methodefficiently utilizes local image information, and therefore it is able to segmentimages with intensity heterogeneity. Moreover, the level set function is no longerrequired to be initialized as a signed distance function, so the comp utation isdeducted apparently. On the other hand, we can improve the computation method toaccelerate the speed of segmentation. So the sparse field method (SFM) isintroduced in our paper. With this method, we only consider the pixels around thezero level set. And the pixels around the zero level set are built as doubly-linked-lists. In the process of segmentation these pixels are added or deletedfrom the doubly-linked-lists dynamically. After the sparse field method is applied,the computation in the iteration progress is dramatically deduced. As a result thesegmentation speed is accelerated apparently.Secondly, in recent years, embedded system is developed fast and applied inmany industrial products because of its advantages of reliability, small powerconsumption and low price. In this paper, we build up the embedded system on theMini2440board and verify the segmentation method we improved.