Texture
It is quite easy to understand what a texture is, though somewhat less easy to define it. For many reasons it is useful to be able to classify textures and to distinguish them from one another. It is also useful to be able to determine the boundaries between different textures, as they often signify the boundaries of real objects. This chapter studies the means of achieving these aims. Look out for:
• basic measures by which textures can be classified—such as regularity, randomness, and directionality.
• problems that arise with “obvious” texture analysis methods, such as autocorrelation.
• the long-standing gray-level co-occurrence matrix method.
• Laws’ method and Ade’s generalization of it.
• alternative approaches to texture analysis such as fractal-based measures and Markov random field models.
• the fact that textures have to be analyzed statistically because of the random element in their construction.
Texture analysis is a core element in the vision repertoire, just as textures are core components of most images. Thus, this topic had to appear somewhere in Part 4 of the book; its location immediately after the chapters on statistical pattern recognition and neural networks appeared to be the most appropriate.