In this chapter, we learned how to take an image pixel and threshold it with its neighboring pixels within a given radius and then perform a binary and integral operation to create an LBP pattern. The LBP pattern is a good example of unsupervised machine learning as we did not train the classifier with the output; instead, we learned how to adjust the parameters of LBP (radius and number of points) to arrive at the correct output. LBP was found to be a very powerful and simple tool for texture classification. However, when the image was non-textured, LBP did not return good results and we learned how to develop an RGB color matching model to match colored non-textured images such as face and foundation color. To create an LBP representation, the image has to be converted into grayscale.
In the next chapter, we will introduce the concept of the integral image by combining...