A pixel gradient-based adaptive interpolation filter for multiple view synthesis

A pixel gradient-based adaptive interpolation filter (PGAIF) is proposed in this paper to solve the over-blurring problem of conventional interpolation filters in applications such as multiple view synthesis, especially when pixels of input pictures follows an irregular pattern. In the PGAIF method, the directional information of input images is utilized, and the coefficients of each filter are adapted with local image structures. A combining kernel function is presented for this proposed filter, and an implementation method is recommended. The selection of parameters in the methods is also studied, which can guide the design of the filter. Simulations of the proposed method demonstrate a significant improvement of visual quality compared with the bi-linear filter.