Abstract: Gait images obtained with complex background at a distance are usually much affected by noise. Gabor features show good performance in the gait recognition work under this situation, but the multiple templates used by some Gabor features based algorithms lead to increased computational complexity. To solve the problem, a new improved Gabor features based gait feature extraction and representation method is proposed in this paper. Valid regions in gait energy image are emphasized, and regions susceptible to noises are suppressed. A basic filter with complementary characteristic in two directions is constructed, and a series of filters are generated by scaling and rotating the basic filter. The filters are used to convolute with improved gait energy image and gait difference image, two feature vector sets are obtained and used to represent the gait subject. The nearest neighbor classifier is adopted to calculate recognition rates on USF database. The comparison result with relevant algorithms confirms the effectiveness of the proposed feature extraction and representation method. The analysis of computational complexity shows that, the proposed algorithm achieves less computational consumption compared with relevant algorithm.