Abstract: To solve the problems that the description of image information is simple and the outline of the objective is easily influenced by background disturbances, an adaptive random walk (RW) image segmentation algorithm is proposed. A texture-based similarity weight expression is given, with the texture features introduced into RW algorithm for the first time to highlight the image structural information. In order to accurately calculate the weight between two adjacent nodes, an adaptive weight expression is proposed, i. e., the proportion of intensity-based and texture-based weights in weight expression will be adaptively calculated according to the image edge density. High-quality segmentation results can be achieved by solving Dirichlet boundary condition. The experiments demonstrates that the proposed algorithm accurately describes image structural information and is more applicable and accurate in comparison with graph cut (GC) and typical RW algorithms.