Evaluation of the results of pansharpening multispectral images

Abstract

When processing digital images obtained by remote sensing of the Earth, various methods are used to increase their resolution. However, in this case, some distortions of a different nature may appear on the images. For example, luminance distortion (color, contrast, sharpness) and geometric (object boundary deformations). Developers of automated image processing systems face the task of choosing from dozens of methods the one that introduces the least visually noticeable distortions, i.e. creates images of the best quality.In this article, the following problem was solved: to determine the functions for assessing the quality of images formed as a result of multispectral satellite image pansharpening. The pansharped image cannot be compared with the template one, since it does not exist. To assess quality of such images, we proposed to use the so-called no-reference evaluation measures.The article briefly describes methods for synthesizing a new high-resolution color image from four images of Earth remote sensing. Functions for calculating quantitative estimates of the quality of the resulting images are discussed. Results of some space image pansharpening by different methods are presented. Graphs of these assessments of image quality are constructed. To evaluate panchromatic fusion results, the following non-reference quality scores are recommended: FISH, LOCC, LOEN, NATU, SHAR, and WAVS. The clearest boundaries and natural colors of objects were demonstrated by the P+XS pansharpening algorithm based on a linear combination of spectral channels.

Keywords

• multispectral image
• panchromatic image
• resolution increase
• pansharpening
• quantitative image quality assessment