Abstract: Due to the overcomplete codebook and the independent coding processing,the similarity of the image is lost between block and block to be encoded．To preserve such similarity information,we proposed the image retrieval algorithm based on Laplacian sparse coding．Given initial sparse coding and calculating the Laplacian matrix,similarity preserving term was incorporated into the objective of sparse coding．We used the feature-sign search algorithm and the golden section line search algorithm to update one by one each coefficient of sparse coding．The experiments show that Laplacian sparse coding can enhance the robustness of sparse coding．Compared with the improved SPM model,the new image retrieval algorithm better improves the retrieval accuracy.

Key words: Sparse coding,Image retrieval,Codebook,Laplacian matrix,Similarity matrix