This paper presents a cross-media relevance model for automatic image annotation and retrieval. The model uses probabilistic methods to predict the likelihood of words appearing in images based on their visual features. The approach involves clustering image regions into blobs and using a training set of annotated images to learn the joint distribution of blobs and words. This allows for automatic image annotation and retrieval based on queries. The model outperforms existing methods such as the Co-occurrence Model and the Translation Model in terms of mean precision and recall. The paper also describes two retrieval models: one based on annotation and another that directly retrieves images by comparing blob probabilities. Experimental results show that the cross-media relevance model significantly improves annotation and retrieval performance. The model is effective in capturing semantic relationships between images and text, and can be used to correct errors in manual annotations. The paper concludes that cross-media relevance models are a powerful tool for image annotation and retrieval, and suggests future research directions in this area.This paper presents a cross-media relevance model for automatic image annotation and retrieval. The model uses probabilistic methods to predict the likelihood of words appearing in images based on their visual features. The approach involves clustering image regions into blobs and using a training set of annotated images to learn the joint distribution of blobs and words. This allows for automatic image annotation and retrieval based on queries. The model outperforms existing methods such as the Co-occurrence Model and the Translation Model in terms of mean precision and recall. The paper also describes two retrieval models: one based on annotation and another that directly retrieves images by comparing blob probabilities. Experimental results show that the cross-media relevance model significantly improves annotation and retrieval performance. The model is effective in capturing semantic relationships between images and text, and can be used to correct errors in manual annotations. The paper concludes that cross-media relevance models are a powerful tool for image annotation and retrieval, and suggests future research directions in this area.