Digital Image Interpolation in Matlab

7 Edge‐Directed Interpolation

The image interpolation methods presented in previous chapters can deliver high visual quality interpolated images. However, no matter how the interpolation is performed in spatial domain or in transform domain, the interpolated images still have different levels of visual artifacts, particularly around the areas of image edges and high frequency texture‐rich regions. The most obvious artifacts are usually observed in the form of blurred edges, blocky image, edge discontinuities, lack of fine details in texture‐rich areas, etc. Different from the task of up‐sampling a one‐dimensional (1D) signal, up‐sampling an image should exploit the fact that high frequency image features, such as edges and textures, are anisotropic in nature. The spectrum of the edges is also asymmetric since the frequency is low along the edge direction and high in the direction across the edge. As discussed in Chapter 4, the interpolation is mostly similar to a low‐pass reconstruction process, which imposes technical difficulties to preserve all the frequency components in the interpolated images, in particularly in preserving the high frequency components of the anisotropic edge spectrum, which results in distorted edges and loss of fine details in the interpolated images.

If we further investigate the causes of interpolation artifacts, we shall notice that the nonparametric interpolation methods introduced in previous chapters interpolate all pixels in the same fashion, ...