Building Computer Vision Projects with OpenCV 4 and C++

Active appearance models and constrained local models

A canonical example of a holistic method is the active appearance model (AAM) from the late '90s, usually attributed to the work of T.F. Cootes (1998). In AAM, the goal is to iteratively match a known face rendering (from the training data) to the target input image, which upon convergence gives the shape, and thus, landmarks. The AAM method and its derivatives were extremely popular, and still take up a fair share of attention. However AAM's successor, CLM methods, have shown far better performance under illumination changes and occlusions, and rapidly took the lead. Mostly attributed to the work of Cristinacce and Cootes (2006) and Saragih et al. (2011), CLM methods model the pixel intensity ...