Interpolation of Multispectral Images Based on Convolution with the Geodesic Distance Kernel and Quality Estimation Using the Structural Similarity Index Criterion

Abstract

The convolution kernel based on the geodesic distance has many advantages because it admits recursive computation and, therefore, fast image processing. Besides, the interpolation quality in some channels increases in the presence of additional image channels with a higher resolution as compared to the interpolated image layer, which is important for solution of the problem of multispectral image interpolation. In this paper, the quality is estimated using several criteria, in particular, the structural similarity index. In the experimental part of the paper, the evident advantage of the proposed interpolation over traditional methods (in particular, the bicubic and bilinear interpolation) is shown.