Focusing keywords to automatically extracted image segments using self-organising maps
In this chapter we consider the problem of keyword focusing. In keyword focusing the input data is a collection of images that are annotated with a given keyword, such as "car"'. The problem is to attribute the annotation to specific parts of the images. There exists plenty of suitable input data readily available for this data mining type of problem. For instance, parts of the pictorial content of the World Wide Web could be considered together with the associated text. We propose an unsupervised approach to the problem. Our technique is based on automatic hierarchical segmentation of the images, followed by statistical correlation of the segments' visual features, represented using multiple Self-Organising Maps. The performed feasibility study experiments demonstrate the potential usefulness of the presented method. In most cases, the results from this data-driven approach agree with the manually defined ground truth for the keyword focusing task. In particular, the algorithm succeeds in selecting the appropriate level of hierarchy among the alternatives available in the segmentation results.