| Textural information is a large part of the huge amount of available digital image data. Although research in texture analysis is being conducted for over three decades, there are still important unanswered questions. One such question is what is the smallest patch from a texture that contains all perceptually relevant textural information in the concerned texture. An answer to this question is important for many fields such as texture synthesis, texture compression, image database retrieval, and 3D vision. Partial answers were proposed in the structural texture analysis field. However, those solutions work on a very narrow range of textures - those featuring high regularity. We aim at giving an answer with a broader applicability scope. We intend to apply methods used in dynamical system theory for time series analysis. Such methods proved successful in studying dynamical system cases that bear resemblance with texture analysis situations. The methods are based on Renyi's generalized entropies. The main challenges in solving the mentioned problem are finding the size and the shape of the smallest texture patch containing the essential textural information and selecting the relevant textural features to be used for texture characterization. Another aspect to be studied is the efficiency and robustness of various Renyi's entropies in solving the mentioned problem. The proposed research builds on the expertise existent in the research team and fits existent trends in the image processing community. The results will be submitted for publication to high standing journals and conferences and the developed algorithms will be made available for the international scientific community through the web as a MATLAB source code and a Java applet. |
