Object Contour Extraction Based On String Theory

By taking into account the high-level image information and the property of contour,the differential equation model that in context of the quantum mechanics and classical mechanics makes the object contour extraction into a simple mathematical problem to research,which is useful to imageanalysis and recognition.In physics,the string theory based on wave model is proposed to unify the two theories which isbased on a particle model.By introducing the open string and closed string,this paper gives the parameter equation underDirichlet boundary condition and two ways to extract contour,which can ensure the integrity of contour curve,so the curve can be zoomed without limitation.In the first method,we establish energy functional model,so the solving minimal value problem can be converted to finding the solution of the corresponding nonlinear equation group,with thediscretemodel.Taking into account the discrete nature of the image data,we use the rank-1 inverse Broydenmethod to avoid calculating the inverse of the coefficient matrix.After setting the appropriate parameters,the optical images,X-ray images can extracted effectively.The ultrasonic images containednot too much noise can also get good result by increasing the number of manual interactive points.However,this method just can get local extremum,which needs the selection of initial points strictly.In the second method,we change the definition of the external potential energy of the contour and add length penalty factor.In order to find out the minimal value of the energy model,the object contour curve is considered as the system that made up of interacted discrete particle series,a response function is designed to simulate the interaction between the particles.In the process of solving the model,we can gradually search the contour curve of satisfying condition,finally extract the object contour satisfying a minimum of the energy functional.Compared to the first method,we can get better contour curves in this three kinds of images.