Feature Extraction for Image Understanding in CPODW

This paper presents two classes of image features, namely relative and absolute features, designed for image content understanding in multimedia content protection by using digital watermarking techniques. Relative features are extracted based on image segmentation. Absolute features, on the other hand, characterize properties such as image edges and principal frequency components. Experiments have been conducted to compare the performance of these features according to the feature selection rules proposed in the generic framework CPODW [1]. These experiments indicate that relative features are more robust in terms of non-content operations, whereas absolute features are more sensitive to image alteration. We also demonstrate that these two types of features should be combined in order to improve the overall performance in practice.