Super-resolution with Adaptive Pixel Weighting Scheme and Its Application to Super-resolved Free-Viewpoint Image Synthesis

The objective of reconstruction-based super-resolution is to construct a high-resolution image from a sequence of low-resolution images. The super-resolution processing requires precise registration of low-resolution images and the registration error degrades the resulting high-resolution image. In order to handle inaccuracies in registration, the robust super-resolution methods have been proposed. We propose an adaptive pixel weighting method which uses pixel weighting variables for the input pixels. The proposed method is the extension of the conventional pixel selection type robust super-resolution. We define the cost function which includes the regularization term for the weighting variables and the cost function is minimized against the weighting variables and the resulting HR image simultaneously. We confirmed the effectiveness of the proposed method with super-resolved free-viewpoint image synthesis.