Image interpolation based on image statistics: a case study

Two novel approaches to image interpolation aimed at video service interworking, layered image coding, and image zooming are discussed. Compared to conventional interpolation filters used for image interpolation, the schemes proposed take advantage of prior knowledge of image statistics to enhance their performance. Both methods, inverse Wiener filtering (IWF) and vector interpolation (VI), use image training sequences to estimate optimal interpolation parameters. While IWF has online capabilities for parameter estimation, VI parameters need to be designed in computer intense offline optimization procedures. Both IWF and VI methods achieve superior performance compared to standard linear interpolation filters in terms of signal-to-noise ratio of the interpolated images, as well as entropy of the residual errors.<
>