In this paper，based on biologic visual information processing，an improved Itti’s visual attention model is proposed for image retrieval.By considering the relation between texture and visual perception，a coarseness map is constructed as a primary visual feature of the new visual attention model.Therefore，by the improved Itti’s model，we can obtain 50 feature maps，and extract the distributions information over the local binary pattern histogram fourier (LBP-HF) of each feature maps to get the high dimensional features.Finally，the locality preserving projections (LPP) is utilized for dimensionality reduction，and the low dimensional feature having both the local geometry and the discriminate information is used for image retrieval.Experimental results show that the proposed approach has good retrieval performance.