Modelování dynamických textur

Abstract

This dissertation thesis deals with a novel dynamic texture model. The proposed toroidal model can represent a vast amount of possible dynamic textures realisation. The proposed model is based on triple toroid-shaped tiles and capable of texture analysis, synthesis, enlargement, and editing. The crucial toroidal property allows synthesizing innitely large output simultaneously in a spatial and temporal domain. Furthermore, the analysis and synthesis parts are entirely separated. This dissertation thesis presents several techniques to enhance the visual quality of results and their potential amount. Particular attention is therefore paid to a dynamic texture inpainting problem, especially inpainting with a dynamic background which is generally recognized as a challenging problem. Afterwards, the dynamic texture similarity the problem is addressed and a novel Fourier transformation based criterion is presented. Subsequently, with proposed criterion utilising and psycho-physical test, the results of stated inpainting method and synthesizing result are validated. Finally, a model capability, particularly its editing abilities, to extend the existing textural BTF model to the dynamic domain is discussed and furthermore illustrated. Consequently, the novel DBTF model for representing a dynamics object is introduced.

This dissertation thesis deals with a novel dynamic texture model. The proposed toroidal model can represent a vast amount of possible dynamic textures realisation. The proposed model is based on triple toroid-shaped tiles and capable of texture analysis, synthesis, enlargement, and editing. The crucial toroidal property allows synthesizing innitely large output simultaneously in a spatial and temporal domain. Furthermore, the analysis and synthesis parts are entirely separated. This dissertation thesis presents several techniques to enhance the visual quality of results and their potential amount. Particular attention is therefore paid to a dynamic texture inpainting problem, especially inpainting with a dynamic background which is generally recognized as a challenging problem. Afterwards, the dynamic texture similarity the problem is addressed and a novel Fourier transformation based criterion is presented. Subsequently, with proposed criterion utilising and psycho-physical test, the results of stated inpainting method and synthesizing result are validated. Finally, a model capability, particularly its editing abilities, to extend the existing textural BTF model to the dynamic domain is discussed and furthermore illustrated. Consequently, the novel DBTF model for representing a dynamics object is introduced.