An Improved Random Walk Algorithm for Image Segmentation
Yujian Li · Computer and Modernization · 2013
An improved random walk algorithm for image segmentation is proposed to solve the problems of heavy computation and complex,a very slow speed with the increase of pixels and unreliable segmentation performance to color image of the traditional random walk algorithm for image segmentation in pixel.First,an improved and optimized watershed transform is used to partition the image into many small homogeneous region pieces.In order to avoid over-segmentation problem of watershed,the image is preprocessed with nonlinear anisotropic diffusion smooth algorithm morphologic processing;secondly,the graph based on region is build on the homogeneous region pieces pre-segmented by watershed transform and used as graph vertexes of random walk algorithm.Then,the target object interested is segmented out following the user guidance to segment the object.Finally,some experimental results are conducted to compare,analyze and evaluate the performance of traditional random walk and the new proposed algorithm.A conclusion and future development is made in the end.