| 摘要 |
The present invention relates to a method for creating a multi-frame-based super-resolution image, which receives a low-resolution image and creates a multi-frame-based super-resolution image using motion vector normalization and edge pattern analysis. The method comprises the steps of (a) selecting, from a low-resolution image, an image at target time and an image at time before and after the target time; (b) expanding, to pixels, the sub pixels of the images (hereinafter, target image) at the target time and at time before and after the target time, and interpolating the target images by applying a 6-tap finite impulse response (FIR) filter (hereinafter, 6-tap filter); (c) estimating motion vectors of block units in the expanded image at time before and after the target time, based on the expanded image at the target time; (d) normalizing the estimated motion vectors by modulo for the extended multiples of the target images; (e) verifying matching points induced by the motion vectors according to edge pattern analysis; and (f) estimating the pixel values of final sub pixels of the expanded images through a one-dimensional kernel estimation formula. The method for creating a super-resolution image as described above can achieve superior objective and subjective performance as compared with a conventional bi-linear interpolation method, a conventional single image-based super-resolution image creation method, and a conventional multi-frame-based super-resolution image creation method. |
