Bachelor Assignment
Image formation modeling of face images from a distance
Type: Bachelor EE
Duration: April 2018 until Juli 2018
Student: Wang, H. (Haichuan, Student B-EE)
Supervisors:
Introduction:
Biometrics is about recognising persons based on their physical properties, behaviour or traces they leave. Examples of biometric modalities are face, fingerprint, iris, voice etc. In 2D face recognition one of the challenges is face recognition at a distance, where the resolution of the facial images can be as low as 10x10 pixels or even lower. At DMB, we have developed several new face recognition methods specically for low resolution, amoung which are so-called mixed resolution face recognition (comparison of low resolution trace material with high resolution reference images) and methods based on deep learning. One of the problems we are confronted with is the lack of low resolution data face bases. Especially for deep learning much data is required. Also we'd like to better understand face image formation in order to improve our methods.
Figure 1: From left to right: face recorded at different distances, real low res face, downsampled
Assignment:
In many publications on low resolution face recognition, the lack of low resolution face data is "solved" by down sampling high resolution facial images. As can be seen in the figure, down sampling high resolution facial images gives images that are of far better quality than real low resolution facial images. The reason for this is that there are many other processes playing a role in the image formation of low resolution face images. For example, if images are recorded at a large distance, the viewing angle is much smaller, resulting in geometric diferences relative to images recorded from closeby. Other effects are noise, atmospheric distortion, compression, optics and others. The aim of this assignment is to develop an accurate image formation model that can be used to synthesize realistic low resolution facial images from high resolution facial images and may be of help in further development of low resolution face recognition.