Image Classifiers for Scene Analysis

The semantic interpretation of natural scenes, generally so obvious and effortless for humans, still remains a challenge in computer vision. We intend to design classifiers able to annotate images with keywords. Firstly, we propose an image representation appropriate for scene description: images are segmented into regions and indexed according to the presence of given region types. Secondly, we propound a classification scheme de- signed to separate images in the descriptor space. This is achieved by combining feature selection and kernel-method-based classification.