Concept-oriented Sample Images Selection

In semantic-based image classification, learning concepts for adding knowledge to the image descriptions is an issue of special interest. This learning increases the capabilities for more “intelligent” image processing. The classifier learns by generalizing specific facts present in a number of design samples. Due to the fact that the learning and classification processes run over image descriptions containing part of the image content, selection of training patterns should take into account relationships among those descriptions. Proposed framework uses unsupervised clustering to support the selection of design samp les and user feedback to refine the classifier model.