Local Patch Active Appearance Model of the Face

This paper discusses the enhancement of the Active Appearance Model (AAM), a classical algorithm in machine learning traditionally applied to facial image processing tasks. The focus is on improving the accuracy and adaptability of AAM for solving complex tasks, including tracking face contours for animation creation. The model is trained on a personalized sample of facial expressions, enabling it to efficiently account for both global and local features of an individual’s face images. The study highlights the limitations of traditional AAMs based on global models of shape and texture in processing local details. The use of localized texture patches and the transition to local piecewise-linear shape models significantly enhance the model’s performance in tasks involving detailed processing of face images. The modified version of AAM demonstrates improved results, confirmed experimentally, and facilitates faster and more accurate tracking of facial features for use in digital animation. The algorithm is applicable for any facial landmarks, but within the scope of this article, it will be considered using the example of lips, as the most challenging region.