Image interpolation using multiresolutional critical point filters with unconstrained boundary extension

Multiresolutional critical point filters (CPF), a set of nonlinear filters, has been shown to be effective in image matching interpolation. In CPF matching, the elastic model defined by the bijectivity conditions can express deformable motions accurately. But the surjection condition constrains that the pixel on the boundary of the source image should be mapped to the pixel with the same location at the destination image. This does not hold in many cases, and will lead to mapping convergence at the image boundary, and thus degrading the quality of interpolated images. To solve this problem, this paper proposes an unconstrained boundary extension (UBE) approach. By UBE, mapping convergence at the image boundary can be released. Experimental results show that the proposed approach can improve the visual quality of the interpolated images, especially at the boundary regions.