Local Configuration Features and Discriminative Learnt Features for Texture Description

Textures, low-level image features around all of us, can be quantified in many ways. One of the most representative methods is the Local Binary Pattern (LBP) operator. In this chapter, two texture description methods inspired by LBP are presented. They are designed in unsupervised (i.e., class labels of texture images are not available) and supervised (i.e., class labels of texture images are available) manner, respectively. First, a linear configuration model is proposed to describe microscopic image structures in an unsupervised manner, which is subsequently combined together with LBPs. This descriptor, denoted as Local Configuration Pattern (LCP), is theoretically verified to be rotation invariant and able to provide discriminative complement to the conventional LBPs. Second, in the case that class labels of training images are available, a supervised model is developed to learn discriminative patterns, which formulates the image description as an integrated three-layered model to estimate optimal pattern subset of interest by simultaneously considering the robustness, discriminative power and representation capability of features. This model can be applied to many LBP variants, such as completed local binary pattern (CLBP) and local ternary pattern (LTP).