Bibliography

(eng): The goal of artwork analyzes is often to detect of pentimenti, retouches, overpaintings, or varnishes in order to understand a painting structure. A common model of a painting used for interpretation of an artwork multimodal dataset is based on its multilayer characteristics. Another possibility how to address an artwork structure is to study an information gain of a particular modality. We have developed a new approach [2] for the information gain extraction and demonstrated its applicability. We present a comparison of four methods for the information separation [4, 1, 3, 2] applied on a multimodal dataset. Their ability to uncover concealed features of paintings will be presented together with their requirements and limitations. The separation limits will be shown using a concept of the intensity correspondence matrix (ICM), which can well describe the correlation and the mutual information. ICM also gives evidence of possibility to achieve an effective signal separation.