|M.Sc Student||Hait Fraenkel Ester|
|Subject||Quality Enhancement of Facial Images Based on|
|Department||Department of Electrical Engineering||Supervisor||Professor Guy Gilboa|
|Full Thesis text|
Image processing algorithms, which have been steadily improving over the past decades with the use of better image priors, have nevertheless started to reach asymptotic improvements in recent years. To overcome the classical processing limits we combine semantic data and registration algorithms. Our aim is to solve image quality enhancement problems, such as denoising, super-resolution and color correction, in the case of facial images. Given today's easily available photography tools, our model assumes prior high-quality data of the person to be processed, but no knowledge of the degradation model.
We use semantically-aware patches, with adaptive size and location regions of coherent structure and context, as our building blocks. Data-driven affinity spaces of facial features, displaying various expression variations, are constructed using a newly-defined affinity measure derived from the non-rigid Demon registration. Its robustness to quality degradation allows accurate matching of low-quality features to similar examples, to obtain a high-quality facial image, while preserving identity, pose and expression. We show how our method significantly enhances image quality, both visually and quantitatively, for the common problem of cellular photography enhancement of dark facial images, for different expressions, poses and identities, relying on only tens of personal priors, and compare it to state-of-the-art denoising, deblurring and super-resolution methods.