Print Email Facebook Twitter 3D Face Recognition: How to make a fast and reliable database and compare the database with 2D+3D facial input? Title 3D Face Recognition: How to make a fast and reliable database and compare the database with 2D+3D facial input? Author Huijbregts, M.A. Stobbe, B. Contributor Mandai, S. (mentor) Faculty Electrical Engineering, Mathematics and Computer Science Department Microelectronics & Computer Engineering Programme Circuits and Systems Date 2013-07-01 Abstract A 3D face recognition algorithm has been developed for the Microsoft Kinect during the final bachelor project at Delft University of Technology in 2013. The aim of the project is to develop a prototype face recognition system. The prototype system has to outperform the existing 2D face recognition system. The main goal is to develop a 3D face recognition system and is divided into three parts. Each part was developed by a group of two students. The subjects were data-acquisition, data-processing and data comparison. Data-comparison is the main topic of this thesis. Nowadays, security is an increasingly important topic in society. 3D face recognition can contribute to make the world a safer place. This thesis is about discovering the new 3D techniques in chapter 4, but first getting familiar with the 2D-world in chapter 3. Our results showed that a 2D system is not accurate to use as a face recognition system. This system is too sensitive for differences in lightning, poses and face expressions. Meanwhile the results achieved with our (proposed) 3D system were quite fast and accurate. The 3D system had five correct matches from the possible six matches. So only one person was not recognized and the process time is 1.19 seconds. This is not tested enough during the period that this thesis was created. It is concluded that 3D face recognition works accurately with geometry and normal map input in combination with the Haar-Walsh Transform and angle-based distance, even though the number of data is not sufficient yet. It even works with only the geometry image as input, both the Haar-Walsh and the Haar transform and the angle-based distance. These two options gave the same result. So as a future task we will increase the number of data to confirm our algorithm. Subject face recognitionmatchingdistance measurementsHaarHaar-Walsh To reference this document use: http://resolver.tudelft.nl/uuid:44ceb68b-7901-4c3e-83c6-53a593e136d6 Part of collection Student theses Document type bachelor thesis Rights (c) 2013 Huijbregts, M.A.Stobbe, B. Files PDF Bsc_thesis_Face_Recogniti ... _input.pdf 2.68 MB Close viewer /islandora/object/uuid:44ceb68b-7901-4c3e-83c6-53a593e136d6/datastream/OBJ/view