Traditional　linear　learning　graph　matching　model　is　easy　to　be　trained　and　can　achieve　a　global　optimal　solution.　However,　this　model　doesn＇t　consider　the　information　of　graph　structure,　thus　limiting　its　matching　accuracy.　To　overcome　this　disadvantage,　we　propose　a　novel　linear　learning　graph　matching　model-edge　feature　based　learning　complete　graph　matching　model　(ELC-GM).　An　edge　feature　is　constructed　from　its　sampling　point　features,　which　are　described　by　an　extension　of　shape　context　with　rotation　invariant　factors.　After　supervised　training　of　ELC-GM,　Kuhn-Munkres　is　used　to　solve　the　edge　match　and　then　Hungarian　decoder　is　applied　to　determine　the　final　point　match.　Experimental　results　show　that　ELC-GM　can　achieve　good　performances　with　improvement　of　accuracy,　even　in　cases　of　deformation　and　noise.　©　2017,　Beijing　China　Science　Journal　Publishing　Co.　Ltd.　All　right　reserved.