Face deformation system based on active shape models ASM

Abstract

This work shows a system oriented to generate a deformation and transformation of human faces between two images by applying Active Shape Models (ASM). The software was developed on Visual Studio and the free OpenCV libraries was used. The first step was to build an ASM model based on the Yao Wei's algorithm. To create the ASM model, the IMM face database, from the Technical University of Denmark, was used to train and build the model by applying an algorithm based on image pyramidal techniques. With the ASM, we create an algorithm to adjust both images to deform and transform one face into another. The deformation is performed by using Delaunay triangulation in the first frame of one image, and then, in a routine of 20 steps, the first face is deformed and averaged with the second face by doing weighted sum. The experimental results show a good performance on the deformation algorithm and the homographic transformations of each triangle in the mesh.