Structure adaptive up-scaling using principal component analysis

Up-scaling video to higher spatial resolutions is a common necessity in current TV systems. Various different up-scaling algorithms have been developed to resize the source video to higher display resolutions while attempting to preserve sharpness. An attractive method in terms of image quality is structure adaptive up-scaling, which typically classifies the pixel structure using adaptive dynamic range coding (ADRC). Yet, a major disadvantage of ADRC is that it lacks memory efficiency as the number of classes can become large. Moreover, as ADRC is simple but not statistically optimal, the accuracy of the up-scaling may be improved further with an alternative classification method. This paper proposes a new classifier, PC-ADRC, which classifies a pixel on the basis of its principal component, rather than on the pixel values directly. We show that the new classification is memory-efficient and that it provides an easy way to reduce the number of classes. Furthermore, with PC-ADRC one achieves a similar but not improved accuracy over ADRC.