The task of automatic image annotation is of great interest because it can play a crucial role in building an effective engine for image retrieval. Assigning descriptive keywords to images allows users to search for images using only text-based queries. Evaluating the performance of an image retrieval engine is different than that of an annotation engine because in retrieval we are only interested in the quality of the first few images associated with a given keyword. Following (Carneiro et al., 2007), we have reported the average retrieval precision over all keywords, as well as just the recalled keywords, for the first 10 retrieved images. It is widely acknowledged that image annotation is an open and very difficult problem in computer vision. Solving this problem at the human level may, perhaps, require that the problem of scene understanding be solved first. However, identif