Leveraging Scene Text Information for Image Interpretation