Remote Sensing Image Fusion

Chapter 3

Quality Assessment of Fusion

3.1 Intro duction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.2 Quality Definition for Pansharpening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

3.2.1 Statistical Quality/Distortion Indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.2.1.1 Indices for scalar valued images . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.2.1.2 Indices for vector valued images . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.2.2 Protocols Established for Pansharpening . . . . . . . . . . . . . . . . . . . . . . . . . 57

3.2.2.1 Wald’s protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3.2.2.2 Zhou’s protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

3.2.2.3 QNR protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.2.2.4 Khan’s protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.2.3 Extension to Hyp erspectral Pansharpening . . . . . . . . . . . . . . . . . . . . . . . 61

3.2.4 Extension to Thermal V-NIR Sharpening . . . . . . . . . . . . . . . . . . . . . . . . 62

3.3 Assessment of Optical and SAR Fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3.4 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.1 Introduction

Quality assessment of remote sensing image fusion has been, and still

is, the object of extensive research activities. Unlike fusion aimed at vi-

sion/detection/recognition purposes, like surveillance or for military applica-

tions, in which quality is mainly related to a subjective perception that may

be embodied by several statistical single-image indices, like contrast, gradi-

ents, entropy, and so on, remote sensing image fusion requires the definition

of more stringent and quantitative measurements that involve both original

images and fusion product images. Since the target of fusion is generally un-

available and thus cannot be used as reference, several protocols of quality

evaluation have been developed to overcome the lack of a reference.

Given two datasets, fusion aims at producing a third dataset that inher-

its some of the properties of its components. The quality of a fusion product

should be related to its consistency with the component datasets according to

the different properties that are inherited. In most cases, one of the datasets

involved in the fusion process features a spectral diversity. The different re-

sponse of the imaged scene in different intervals of wavelengths is responsible

for the spectral information. After fusion, at least in principle, it should be

51