Face and eye categorization and detection

Title Face and eye categorization and detection
Summary To build a new database of face and eye images of different species and to evaluate holistic and local detection algorithms
Author Zhao Cui, Albert Hoxha
Supervisor Fernando Alonso-Fernandez
Level Master
Status Ongoing

Several algorithms have been developed for face detection, mostly based in scanning the image with sub-windows in a raster-like fashion, and classifying regions of the image as face/non-face. Face is modeled holistically and detected globally, thus failing in case of perturbations like arbitrary poses or occlusion. We propose in this project the detection of faces by local landmarks instead (starting from eyes, then to nose, mouth, etc.), which will be used to build up a complete face model. The target here will not only be human faces and eyes, but of other animal species too (dogs, cats, etc.)

Firstly, we will build a database of face and eyes from images available on the Internet. Images will be classified according to relevant information such as distance between eyes (i.e. scale), view angle (pose), animal specie, etc. In the second part of the project, we will evaluate existing methods of holistic face tracking (available online) on the new database, as well as our own detection method based in local landmarks developed at Halmstad University. Contributions of the proposed project are relevant to a range of applications from social networking and mobile smartphones  to video surveillance and forensic investigation.

Solid knowledge of image analysis and a good command of Matlab is a requisite