Scale-Space Mutual Information for Textural-Patterns Characterization

The essence of image texture is typically understood by two aspects. First, within a texture-pattern there is a significant variation in intensity values between nearby pixels. Second, texture is a homogenous property at some spatial scale larger than the spatial resolution of the image. Motivated by the essential aspects of image texture, this paper proposes a novel methodology that combines the use of scale-space and mutual information to characterize textural-patterns. Scale-space offers the mechanism for a multi-scale representation of an image, which will be used to address the scale aspect of texture. On the other hand, mutual information provides a measure to quantify the dependency relationship across the scale-space. It has been found that the proposed methodology has the potential to capture different properties of texture such as periodicity, scale, fineness, coarseness, and spatial extent or size. Practical examples are provided to demonstrate the applicability of the proposed methodology.