Research On Network Structure Image Object Segmentation Under Biased Random Walk Algorithms

The network structure in the image generally refers to a set by the line-like structures that are criss-crossed and interconnected in a two-dimensional image,and is widely existed in the fields of medicine,remote sensing,and microscopy image.The automatic extraction of network structure targets is of great significance to various research fields such as medical automatic diagnosis and geographic map drawing.One of the difficulties of the research of network structure is the problem of the break region in the network structure.At present,the mainstream method of connecting the fracture often changes the shape of the original network structure.Considering the group trend of random walk algorithm and the comprehensive advantage of accuracy by individual walker,this thesis takes the biased random walk as the basic theory,uses the physical gravity model as the walker-driven model,to connect the break region in the segmentation of network structure.The main contributions of this thesis include:1)The gravitational biased random walk model is proposed to study the network structure.Inspired by the gravitation theory of physics,the driving model of the walker is adjusted to be the gravitational model,that is,the seed point in the range has a certain gravitational effect on the walker.The effect is to increase the transing probability in the direction of broken region of the individual walker.Under the combined effect of group walkers,the macroscopic walking path is in the form of a network structure and probabilistically connects the broken region.Experiments show that the proposed algorithm has the corresponding connection ability for common break problems,and the original shape of the network structure is guaranteed compared with other algorithms.2)Adjusting the movement of the walker from the driving force model and the motion model,so that the moving path is closer to the linear branch of the network structure object.(1)The connection of the break is mainly based on the gravitational effect of the distant seed point.Therefore,the driving force is adjusted to the far trend force.The magnitude of the gravitation is proportional to the distance,and the normal force balance state is reached near the center line of the network structure.The walker oscillates near the center line to the linear path walk;(2)the center line can be used as the constraint information of the walker,enhancing the transition probability of direction that the walker closing to the center line,and suppressing the transition probability of direction that the walker fleeing from the center line,under the integrated of the group walker,the macro walking path appears as a linear path.The experiment proves that the proposed measure can drive the walker to achieve the linear path walking state.The linear path walk model not only reduces the time complexity of the algorithm,but also has a certain connection ability for some complex break situations.