Image Processing Reference
In-Depth Information
Chapter 9
Performance Assessment of Fusion Techniques
9.1 Introduction
A quantitative analysis enables understanding of various pros and cons of a fusion
technique. A human observer can judge the performance of a fusion technique in
terms of the visual quality of the fused image. However, this approach is completely
subjective, and thus may vary for different observers. Different subjects can perceive
the same image differently, and their perception may be based on several psycho-
visual factors rather than an objective assessment of the scene.
We have discussed in Chap.
2
the problems associated with the subjective
assessment- it is both expensive and time consuming. Additionally, the assessment
is likely to vary from person to person, and hence, it is not consistent. In order to
obtain a unique assessment score, the image may be subjected to a large number of
observers, and their individual scores are statistically averaged. This whole process,
however, turns out to be tedious, time-consuming, and yet not very accurate. An
objective assessment of the image quality alleviates most of these problems associ-
ated with the subjective quality assessment. In this process of analyzing the quality
of an image, several performance measures which can be calculated either from the
image alone, or with reference to some other image are employed. These measures
can be computed without any human intervention, and they do not get affected by
any psycho-visual, or individual differences. An objective assessment, thus, provides
a consistent outcome which facilitates comparison of different images. These pro-
cedures can also be very easily automated as they do not require any inputs from
the human analyst. The objective performance measures bring automation and uni-
formity to the process of assessment. Also, these measures may be used to optimize
the performance of image and video processing systems by integrating their formal
expressions into the system itself [184]. However, what makes this problem difficult
is the fact that there is no universally accepted solution for the quality assessment
of images in remote sensing. In case of hyperspectral image fusion the performance
evaluation faces following challenges:
