Abstract: As compared to a pixels’regular neighborhood system,the morphology and distri-bution of segments are irregular,which makes the quantity of neighborhood cliques in RAG-MRF unpredictable,and therefore the quantification of segments’orientation features be-comes impossible.Aimed at the disadvantages of current segments-based random field in ori-entation feature,we propose a random field model based on segments spatial integrated adja-cent potential energy analysis,a new neighborhood potential energy function based on statis-tic was used,and according spatial adjacent feature indices for segments was designed,andthe results of classify or recognition was improved.