Research On Invariant Feature For Texture

Texture is an important component of a lot of visual information. Understanding visualtextures plays an important role in many vision tasks such as image and scene classification,video understanding, visual retrieval and image-guided diagnosis. Texture analysis is adifficult problem. Despite decades of research efforts on texture modeling, it remains achallenging problem. Obtaining robust texture descriptor plays an important role in textureclassification. The problem called by geometrical and photometric variations includingnon-rigid surface deformation, viewpoint changes, illumination variation, rotation, scalingand etc, is difficult to solve in the computation of texture descriptor. Robust texture descriptornot only has invariability to these variations, but also has high discriminability. Some existingtexture analysis methods have achieved good results on some texture dataset. But there is nota general method which can be used to all texture dataset and achieves excellent performance.So there is need for research on more general and accurate descriptor.In this paper, we proposed a new texture descriptor, the wavelet-based multi-fractalspectra, for both static and dynamic textures. The new descriptor is built on the wavelet-basedspatial-frequency analysis on two complementary wavelet pyramids: the standard multi-scaleone and the so-called wavelet leader one. The introduced wavelet pyramids essentially capturethe local responses of textures in multiple high-pass channels in a multi-scale andmulti-orientation fashion, in which there exists a strong power-law relationship for naturalimages. Such a power law relationship is characterized by the so-called multi-fractal analysis.In addition, two more techniques, scale normalization and multi-orientation image fusion, areintroduced to further improve the robustness of the proposed descriptor. Combining thesetechniques, the proposed descriptor enjoys both high discriminative power and robustnessagainst many environmental changes.We applied the descriptor for classifying both static and dynamic textures. For dynamictexture classification, this texture descriptor needs to be modified a little. In evaluations usingseveral public benchmark datasets, our method has demonstrated excellent performance incomparison with state-of-art approaches.