The Investigation Of Images Contour Extracting Arithmetic Based On Neural Network

With the fast development of computer technology, the digital image processing technique has been taken seriously broadly in a lot of application fields and there are many significant developing accomplishments. The images contour extracting technology is a hot spot of images processing technology research. It has been proved that there are many extensive applications and investigations about images contour extracting technique in fields such as aviation, industry and physic that has created substantive social value and wealth.Active contour model (snake) is a kind of important contour extracting technology,grey images contour is extracted by energy function minimizing that adjust image local character and stretch changing of template,but these arithmetics exist many limitation. Along with artificial neural networks integrate with images contour extracting technology, investigators of contour extracting technology put forward a few arithmetics about grey images contour extracting based on neural network.Now,self-organization mapping algorithm (SOM) is a main way of images contour extracting technologies.This article introduces relevance model of competitive neural networks and puts forward a novel arithmetic of extracting grey images contour, which is called Batch-SOM based on SOM and Snake. At first, characteristic points of images contours are obtained by Edge-Map, contour is changed in local region along with grads and grey by characteristic points that are obtained by Edge-Map technique.At the same time,Because of characteristic points, contour can not traversing weak or broken object edge. Arithmetic model is different from conventional one and needs not an explicit energy function(based on grey or grads) to control contour moving.The characteristic vectors of Arithmetic are come from points of edge that are obtained by Edge-Map technique. This arithmetic adopts grey and grads in local region to control contour moving. BSOM adopts two important ideas to modify neural cell weight. First, when neuron approaches object edge, it is necessary to move to the normal or the opposite direction of its movement, so the value of gradient vector at new location is greater than one of the current location. If the gradient vector is reduced, the neuron does not allow to move a new location, so even if the weak edge or the broken edge, convergence is also guaranteed. Second, if the neuron is in the same region, it moves towards an appropriate feature point.The algorithm uses neuron's own weight as a new control point's position to strengthen the profile.The simulation results validate validity of this arithmetic. Results show that the convergent rate of BSOM is very quick and the effect of extracted image contours is better.