Enhancing latent semantic analysis video object retrieval with structural information