3D image registration

In this thesis, registration of 3D face images is studied. Iterative closest point algorithm that is commonly used is prefered for 3D face image registration. ICP is an effective method that registers generally all 3D objects. At the result of this algorithm, the calculated distance gives us information on the similarity of images and can be used for face recognition. In this study we register the images with different scenario and the results are used for successful face recognition. But ICP algorithm has some problems that it runs slowly and it converges local minimum because of initial condition. For solving these problems some pre-processing applied and succesful registration has done. Two approaches have proposed for solving the initial condition problem. Different angled images decrease the recognition rate. First approach is applying some initial registration to the image for finding the best registration. Second approach is finding the symmetrical plane of face and use this plane to adjust the face direction. K-D tree algorithm that seperates the 3D space to lower spaces and looks for the closest point in this space has applied to solve calculation load problem. In the experiment section, all face, nose-eye region, nose region and profile region are use as test group. 18 faces are used for face recognition and for full face 94,4%, for nose-eye region 94,4%, for nose region 88,8%, for profile 88,8% success achieved. For region that not contains specific points like cheek has 33,3% success achieved.