Due to the recognition rate of single character is low,the method of multi-feature fusion was proposed.BP_Adaboost neural network(BP_ANN)was first used to recognize the B-scan ultrasonic image of cirrhotic liver.Gray level co-occurrence matrix(GLCM)and gray level difference statistics(GLDS)were introduced in this paper.In order to improve the objectivity of the experimental results,uniform local binary pattern(U_LBP)was also applied.The texture features were extracted by any combination of these three methods.Then the feature which was extracted by above combination was input to BP_ANN.It was shown that the combination of GLCM and GLDS was better than any others in this experiment,and the recognition rate was 97%.The design of BP_Adaboost network and the determination of neurons in hidden layer were also discussed.