In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
385 
Band n 
jeand-l 
Bandi 
I 
Pan-sharpening f-* 
; r 
Band n sharpened 
[Barici 2. stiMBfened 
Band 1 sharpened 
Part-sharpening 
Phase congruency 
gg 
Phase congruency 
(£ß 
Spatial consistency 
assessment using 
Phase Congruency 
PC map (Band n) 
[ PC mao (Band 2) 
PC map {Band 1) 
' 
ZNCC 
Band 1 - Spatial consistency 
Band 2 - Spatial consistency 
Band n - Spatial consistency 
Figure 1. Diagram of spatial consistency assessment using 
phase congruency 
spectral consistency assessment (SSIM, ERGAS, and SAM 
measures). Spatial consistency was assessed using Zero mean 
normalised cross-correlation coefficient (CORR), High Pass 
Correlation Coefficient (HPCC) (Zhou, 1998), SSIM, ERGAS, 
and using Phase Congruency (Zero mean normalized cross 
correlation metric) (PC ZNCC). The assessment functions were 
implemented in IDL, while the original Matlab code was used 
for calculation of Phase Congruency (www.csse.uwa.edu.au/ 
~pk/Research/MatlabFns/). 
One example of quantitative assessment of IKONOS urban 
subscene (Athens, panchromatic image size is 4000x4000) is 
presented in Table 1: spectral consistency (SSIM, ERGAS, 
SAM) and spatial consistency assessment (SSIM PAN, ERGAS 
PAN, CORR PAN, HPCC, PC ZNCC). The SSIM PAN, 
ERGAS PAN, CORR PAN are the measures notation for spatial 
consistency assessment (fused image is compared with 
corresponding panchromatic image). Mean values of the 
measures are calculated over the assessed spectral channels. 
The results of quantitative assessment during the second 
assessment setup are presented in Table 2. The dependencies of 
the measures on the quality of the fused images are presented in 
Figures 2 and 3. The characteristics of the resulting images are 
dependent on the GIF-2 hf parameter. Assessment of the pan- 
sharpened images with different quality (GIF-2 method, 
parameter variation) results in different scores and allows to 
illustrate trends of the measures. 
4. RESULTS AND DISCUSSION 
One of important questions during this investigation was: does 
the assessment using PC has the same trend with the other 
measures? The results produced by assessment measures were 
analysed for similarity in trend. 
During the second assessment setup, the same images were 
pansharpened by GIF-2 method with different parameters. GIF- 
2 has a parameter hf which varies in the range [0, 1] and 
controls proportionality (0%-100%) of high-frequency image 
data to be added to low-resolution spectral image. The high- 
frequency information is extracted using Butterworth filtering. 
The higher the value, the wider the Butterworth filter width and 
the more high frequency data is added. Variation of this 
parameter allows to create fused images with desired quality: 
the more high-frequency data is added, the higher spatial and 
lower spectral consistency, and vice versa. Three different 
values were taken for the parameter (0.9, 0.75, and 0.5, i.e. 
90%, 75%, and 50%, respectively) and three fused images were 
produced. These created images are used for estimation of the 
trend between the measures and phase congruency spatial 
consistency assessment. 
The PC ZNCC and SSIM PAN, ERGAS PAN, CORR PAN, 
HPCC illustrate higher spatial consistency produced by the 
IHS, PCA, GIF-1 and GIF-2 methods. This agrees with the 
well-known fact that the IHS, PCA and GIF methods produce 
the best spatially-consistent results with some loss of spectral 
consistency. Here the PC ZNCC illustrates similar results 
comparing with other measures on spatial consistency (Table 
1). For the ATWT fusion, the PC ZNCC and SSIM PAN, 
ERGAS PAN, CORR, HPCC illustrate loss of spatial 
consistency and the highest spectral consistency (SSIM and 
ERGAS). PC-based metric resulted in the lowest value on 
spatial consistency, which correlates with the knowledge about 
the fusion result. GIF-1 and GIF-2 methods provided a 
compromise between the spectral (SSIM, SAM) and spatial 
consistency (PC ZNCC and SSIM PAN, ERGAS PAN, HPCC, 
together). 
High resolution IKONOS multispectral images were used for 
fusion and assessment. The images were acquired in the areas 
of Athens (27 July 2004, 08:46 GMT) and Munich (15 July 
2005, 10:28 GMT) cities. Full spectral image data (four spectral 
bands: blue colour range, green colour range, red colour range, 
NIR range) was used for pan-sharpening and assessment 
experiments. Sub scenes (panchromatic image size is 
4000x4000) were used in the experiments. 
In the first and second setups the pan-sharpened images were 
assessed for spectral and spatial consistency using standard 
widely used assessment measures. Wald's protocol was used for 
Highest score of SAM for GIF-2 method (Table 1) was caused 
by characteristics of the General Image Fusion (GIF) method, 
which provides a good compromise between the spatial and 
spectral consistency. For this particular case, the GIF method 
resulted in good spectral consistency (according to SAM 
measure) with acceptable spatial consistency. 
Table 1 illustrates better values of ERGAS PAN for ATWT 
(3.78) than for IHS (11.17). The opposite trend is shown by the 
PC ZNCC and CORR PAN, HPCC. Such results may originate 
from instability of the MSE estimator (Wang, 2009) in ERGAS 
measure. Also SSIM PAN illustrated low spatial consistency of 
the IHS fusion (SSIM PAN (mean) equals to 0.6314). This