Based On Asm And Aam For Facial Features Localization And Matching Algorithm Research

The Face-Recognition(FR) technology is of so many advantages like non-intrusion, simple equipment and so on, which makes it get more and more attention recently. However, due to effects of facial expression, light and the complexity of face structure, FR technology still can't comfort us. FR systems can generally be divided into three phases: detect face, locate and extract features and classify. Considering that the second stage is the basic phase of FR system and its precision affects the final performance most, the research on locating and extracting features is very important.This paper will base on the Active Shape Model and Active Appearance Model, the outstanding deformable models. Firstly, it analyzed the idea of the two models and some important steps, especially for the locating of key points and texture matching. Secondly, it compared the search strategies of each model, including advantages and disadvantages, and then proposed an improved search algorithm which takes both global and local features into consideration. It analyzed the integrating progress by introducing image energies, and chose the shape parameter as the integrating point, and then made the integrating weights vary according to the searching progress. Finally, some trivial problems in each stage during building a model and the robustness of these models were discussed, and it also introduced an approach of auxiliary labeling tools.Experiments showed that performance of the improved search strategy was good. It won't fall into local texture and could attain more precise point positions comparing to original ones. These virtues made it more stable and capable of providing better data for proceeding processes.