Face extraction in mobile phones
Loo, Zing Zai
Date of Issue2016-04-19
School of Computer Engineering
Face detection and face recognition are very much discussed topics these days. However, there is an intermediate process often being forgotten, commonly referred to as face extraction. Face extraction is a crucial step performed before faces can be recognised accurately. It also open doors to more advanced face-related applications to be developed, such as face expression analysis, age estimation and 3D face modelling. Face extraction requires concepts from face detection, image processing and segmentation. They are a few of many challenging problems in computer vision and graphics. Image segmentation is considered to be challenging because computers do not know what to segment, which parts of an image are the objects of interest and which parts of the image are the background. Besides this, image processing requires a lot of computational power. Despite having numerous image segmentation applications in the market today, most applications demand users to define foreground (object of interest) and background regions before segmentation could be performed. This step is troublesome and tedious for users. The ultimate goal of this project is to reduce user interaction between face extraction applications. Hence, it was proposed to implement a face extraction application on a mobile device with minimal user interaction so that users can achieve an extracted face image within 2 steps. The 2 steps involve selecting the image and choosing an algorithm for the extraction. Users would be given three extraction algorithm choices, GrabCut, Geodesic and Advanced Geodesic. GrabCut offers a fairly complete and accurate extracted image but compromise on speed while Geodesic extracts faces extremely quickly but compromise on accuracy. The final algorithm, Advanced Geodesic strikes a balance between accuracy and speed by taking facial properties into consideration. All in all, the decision is still based on the users’ needs to choose an extraction algorithm that is most suitable for themselves. Face recognition was introduced into the application to wrap up the entire face processing application. It aims to give average users a better purpose and experience. Face recognition recognises the extracted face before returning a name and the confidence level for the recognition. Text-to-speech function was also implemented to increase interactivity between users and the application.
Final Year Project (FYP)
Nanyang Technological University