The Research On Facial Feature Point Localization And Its Application To Caricature Software System

A face is very familiar with us and it is the object which we see most frequently in our daily lives. Everyone has some interests in faces. In academic fields, the face is treated in various fields such as anthropology, psychology, engineering and so on.Facial Caricatures sometimes can impress people more than original photographs. It represents features of individual face compactly and effectively. By extracting and emphasizing facial features of an individual person from an input photograph, it is possible to draw a facial caricature automatically by computer. Most conventional approaches are based a method which calculates the differences between an input face and an average face, and then exaggerate them by the extrapolation technique. In most conventional approaches, the shapes of facial parts and their arrangement are treated together. In one another method, eigenspaces are calculated for shape of each facial part and arrangement of facial parts, respectively. The exaggeration process is performed through each eigenspace independently. Thus, this method possesses the high flexibility in drawing facial caricatures. But still there are disadvantages in this method. One is that it can only process the input image without facial expressions, rotation. And also the accuracy is also not enough. So when put these result images eigen data into our facial caricature software, the output will be not ideal.In our work, we aim to improve the automatic caricature system in by using Active Appearance Model based on Angle-Judgment and Local Binary Pattern (LBP-AAM) to locate facial feature points and then apply it to caricature software system. The AAM is one generative parametric model of objects that describes both shape and appearance variations. These variations are represented by a linear model such as Principal Component Analysis (PCA), which finds a subspace preserving the maximum variance of a given set of data. In our work, for model instance generation, we mark key landmark points manually by using FaceFit tool and obtain three types of model with different rotation. Then we segment the test face into four regions and apply LBP to judge the rotation of the test face, select proper model instance to fit the test face and extract facial feature points automatically. Finally, we divide these feature points into nine groups according to the principle of caricature drawing of caricature software system. We test our new system and the results show that this system can extract facial feature points from facial image with expressions and rotation and generate facial caricature by caricature software system successfully. Experimental results prove that our method increased the fitting accuracy rate by about 27%and the time consumption was decreased by about 9% comparing with standard AAM method.