Texture Discrimination by Local Morphological Multifractal Signatures

Both the fractal dimension (FD) and the multifractal dimensions (MFD) have been widely used to describe natural textures in image processing community. However, due to the essential difference between the fractal reality of digital images and the mathematical fractal model, most FD/MFD estimation algorithms intrinsically produce less accurate results. In this paper, the idea of fractal signature is adopted and extended to the morphological multifractal estimation. As a result, a novel texture descriptor, namely the local morphological multifractal signatures (LMMS), is proposed to characterize the local scaling property of textured images. The LMMS depict the behavior of the morphological MFD over a wide range of spatial scales. The proposed LMMS feature, together with the fractal signature and the morphological MFD, has been applied to the discrimination of Brodatz textures. The comparison results demonstrate that our LMMS feature can differentiate natural textures more effectively