Entropy based Saliency Maps for Object Recognition

Object identification from local information has recently been investigated with respect to its potential for integration and robust recognition. In contrast to existing approaches, we do not use generic interest operators but select regions of interest from top-down information, i.e., with respect to object recognition. Discriminative regions are determined from the information content in the local appearance patterns (imagettes) and consequently enable to model sparse object representation and attention based recognition using decision trees. Recognition performance from single imagettes dramatically increased considering only discriminative patterns. Evaluation of complete image analysis under various degrees of partial occlusion and image noise resulted in highly robust recognition even in the presence of severe occlusion and noise effects.