Image interpolation using consistent neighborhood correlations
R. Sher, Moshe Porat · International Conference on Systems · 2005
We present a new approach to image interpolation using consistent relationships between adjacent pixels in an image. In the first stage, the localized relationships are learned from the input image. In the second stage, the relationships and the concluded governing rules are used to build an interpolated image. Our method is compared with other interpolation methods such as bilinear, bi-cubic and spline for enlargement of images by multiplications of 2. The results indicate significant reduction in the blockiness and smoothing effects compared to existing methods. We present additional results for related applications of audio signals and for color imaging.