Development of nonlinear filter bank system for real-time beautification of facial video using GPGPU

Abstract

A nonlinear filter bank named as an ∈-filter bank is implemented for real-time processing of video in order to make the skin in human faces look beautified. General-purpose computing on graphics processing units (GPGPU) is utilized for this real-time implementation. GPGPU has quite high computational power, and the facial beautification system using the ∈-filter bank is efficiently implemented in real time, although the amount of the computation of this system is large. A new language is developed to design and edit the ∈-filter bank, which enables this system to be edited conveniently by trial and error, reflecting the user's request concerning the smoothness of skin. Skin detection is also applied to limit the area to be processed within the face area, so that the other areas than the face, such as hair and background, are not influenced by the system. Experimental results verify its high performance.