An object approach to the assessment of the spatial quality of pan-sharpened remote sensing images

A correct assessment of pan-sharpened remote sensing imagery (fusion of panchromatic and multispectral images) enables the improvement and ranking of pan sharpening algorithms. For this purpose, numerous quality indicators have been published in the literature; however, there is no consensus on their performance. In general, the pixel is used as a reference for the computation of these indicators and the result is a unique overall value that hardly represents the perceived quality by a human. In this paper, a supra-pixel approach to spatial quality assessment is addressed. Thus a methodology is proposed to exploit the differences among image objects (segments) to generate a spatial quality map. A database of eight fused images and their sources images has been created to test the proposed methodology. The pansharpening process was conducted using state of art fusion algorithms. Considering that usual approaches to quality assessment operate with a single scalar value, a score of the overall spatial quality is derived from this new measurement schema and compared it with other spatial quality measure, the frequency comparison index.