Abstract: This paper proposes a scene recognition approach for mobile robot localization.The multi-channel Gabor filters are used to extract the global texture features of the scene images which are associated with the corresponding locations,and then these texture features are fed back to support the vector machine classifier to determine the logical location of the robot.The algorithm has been tested on the autonomous mobile robot CASIA-I designed and developed by us.The experiment results indicate that the algorithm can reach up to a correct localiztion rate of 91.11%,is robust to the various illumination and contrast,and satisfies the real-time localization demand of the mobile robot.