Improved explicit shape regression face alignment algorithm

Face alignment, which is an important step in face recognition, face beautification and many other face-related visual tasks, is to accurately locate the key points with semantic features in the face image. Explicit shape regression is a representative effective face alignment algorithm based on cascade regression, which can deal with lots of complex cases, but still has some shortcomings. An improved ESR algorithm is proposed in this paper, which consists of two parts. In one part, the first derivative of Gaussian filter-based gradient difference features is extracted to represent the facial appearance, and a random regression forest is learned to predict initial face shapes. In the other part, the predicted initial shape is taken as the initial shape of the ESR algorithm. Experiments show that the improved ESR algorithm has higher accuracy than the original ESR algorithm.