User driven systems to bridge the semantic gap

In this tutorial relevant development in user-driven image annotation and retrieval is reviewed. This includes descriptive learning models starting from empirical parameter adaptation to approaches considering optimisations and complex parametric as well as non-parametric distributions. The review also includes discriminative learning models focusing on estimating the boundaries between classes rather then exact class distribution. A new approach to infer semantic concepts in images is also described. This approach draws on several important ideas including multi-feature space, learning technique pertaining to user provided relevance and an object based modelling to link semantic terms and visual objects.