Since the color infrared image truly sampled at 1 m is available, the correlation between this original image and the fused image can be calculated directly.
Figure 6: Correlation between original image sampled at 1 m
and image fused by HSV transformation.
In Figure 6 the correlation between an original image with a resolution of 1 m and the fusion-derived image is shown for the bands Near Infrared, Red and Green. The fusion method applied here was the HSV transformation. Each pixel defined by its spatial coordinates has two reflectance values, one in the original and one in the fused image, so every point in the scatterplot represents one pixel. If both images were identical, all the points would be located on the straight line which is also plotted for orientation.
While the correlation between both images is very good in the red and green bands, the near infrared seems to be less correlated. The graphs in Figure 7 show the correlations of the same scene, but here the fused image was obtained by the relative contributions method.
Figure 7: Correlation between original image sampled at 1 m
and image fused by relative contributions.
The following table contains the correlation coefficients corresponding to the graphs in Figures 6 and 7:
| Correlation coefficient | |||
| Fusion method | Near infrared | Red | Green |
| HSV transformation | 0.82 | 0.96 | 0.97 |
| Relative contributions | 0.96 | 0.98 | 0.98 |
The reason for the rather low correlation coefficients caused by the HSV transformation is probably due to the way the ``Value'' image is produced. The HSV transformation as it is described by [Foley et al. (1995)] assigns the maximum brightness of the red, green and blue channel to the ``Value''. Therefore, the ``Value'' does not correspond directly to the panchromatic intensity which is an average of the three bands. In this regard it becomes questionable to replace the ``Value'' image with the panchromatic image.