An improved GMM-based method for supervised semantic image annotation

Automatic image annotation is the key to semantic-based image retrieval. In this paper we formulate image annotation as a supervised multi-class labeling problem. The relationship between low-level visual features and semantic concepts is found by supervised Bayesian learning. Color and texture features form two separate vectors, for which two independent Gaussian mixture models (GMM) are estimated from the training set as class densities using the EM algorithm combined with a denoising technique. Two posterior probabilities are calculated, and both their ranks among different concepts are used to determine the labels for the image to be annotated. The emphasis on different low-level features is balanced. Better annotation performance is obtained compared to method that treats color and texture as one feature vector.