Abstract
Super resolution is a basic operation in image processing, involving how to increase resolution of an image with minimum damage to its visual quality. A great deal of work has been done recently on developing super resolution methods for better performance for either pixel-based or patch-based. Such methods are essential for reusing low-resolution contents like old low-resolution pictures or classical movies. This paper proposes a new super resolution method, using the framework of bilateral filtering but adopting special filters instead. Given a low-resolution image, we first gather two kind of required filters. The first kind is Jinc filter defined by the Jinc function in spatial domain, and the other kind is Game filter determined by an evolutionary game in feature domain. Jinc filter is powerful but preserves only spatial similarity when used in filtering. By regarding each unknown high-resolution pixel as a player, the evolutionary game in feature domain further achieves an evolutionary convergent state to obtain the Game filter, hence preserves feature similarity. Combining these two kind of filters, we then perform the bilateral filtering process over the low-resolution image to obtain its enlarged version as result. Experimental results show that the proposed method could keep higher PSNR value than common methods while getting rid of most ringing effect. As the future work, an interesting direction to explore is creating a new measurement, which is both subjective and objective, by mixing human vision and computer vision.