DETECTION OF FACE FEATURE FOR THE REAL TIME STATE RECOGNITION

摘要

The most important problem of the face recognition system is that of accurate extraction of the face area. The methods widely used in the existing researches are methods using edge information, methods using color information, principal component analysis using PCA (Principle Component Analysis), template matching using template, neural network, ASM. (Active Shape Model), AAM (Active Appearance Model), EBGM (Elastic Bunch Graph Model) and the method using.;Among them, the template matching method mainly uses a gray image, and a part or all of each part of the face is formed into one or several templates to search for a part similar to this template in the whole image, and the face. Is the way to extract This method has the advantage that it is not significantly influenced by the background color or race, etc. compared with the method using color information or PCA, but has a disadvantage of long operation time. In addition, the disadvantage of using ASM, AAM, and EBGM is that it is sensitive to the initial condition in addition to the long computation time, similar to the disadvantage of using template matching.;An object of the present invention is to detect the eyes, nose, mouth, and eyebrows, which are the main feature areas of the face, as a previous step for face recognition. A deformable template matching method is used as a method for detecting a feature region robustly to the change in the geometric position and size of the feature region. In order to reduce the computation time, which is pointed out as the biggest disadvantage of the template matching method, the search time on multiple resolutions is reduced. In addition, the present invention went through a normalization process optimized for a method for accurately finding the feature region of the face even with changing environmental and lighting factors and geometric factors such as rotation of the face image.;The proposed algorithm has the advantage of being able to detect features in a shorter time without being sensitive to environmental change and form change than the conventional method.;Face Recognition, Deformable Template Matching, Multiple Resolutions, Optimized Normalization