In recent years face recognition became one of the most prominent and successful areas of image analysis and understanding. For the purpose of this thesis two applications were developed for identity verification using face images. The first application compares two images from either the same or different person. The second application is used as a smart login into a system, where the image is taken by camera and then compared with a specific photo of a user.
Face recognition is roughly divided into six steps: image capturing, face localization and allignment, image correction (rotation, scaling, fixing illumination), representation, feature extraction and projection into subspace and matching computation. In the theoretical part of the thesis we described the procedures used by the algorithm for training a new model and the procedures used for decision making and for computation of matching between images.
Using OpenBR library and its two versions of the 4SF algorithm we trained two models by which the evaluation on ten different datasets was conducted. In addition, we conducted the evaluation of the existing model, trained by the OpenBR library authors. The model, which has achieved the best results during the evaluation, has been used in both applications.
|
