Hello visitor, its been a while since my last post. Starting from now, I will try to post my blog in English language. This is hard for me because I neither English nor Americans. I am Indonesian. I hope you can understand if my grammar or structure maybe very confusing. Please pardon me. I still have to learn. Ok back to the topic, few days ago, I had downloaded the trial/demo version of Leica Photogrammetry Suite version 9.2 from their FTP ( After exploring some new modules, I get excite with a new image fusion/pan sharpening algorithm that as far I know, not implemented in previous IMAGINE or LPS version. The algorithm was developed by Mr Manfred Ehlers from Onasbruck University Germany, and it is called Ehlers fusion. I have tried to use it to fuse quickbird 16 BIT sample data from digitalglobe (multispectral and panchromatic), and the result is awesome. Comparison between original dataset and sharpening result indicated that this algorithm can maintain spectral integrity from original multispectral dataset without losing spatial details from panchromatic dataset. Looking the result, I have idea to compare this algorithm with Mr Yun Zhang PANSHARP algorithm that now implemented in PCI Geomatica (, and Gram-schmidt Spectral Sharpening that now implemented in ITT ENVI ( The reason why I chosen PANSHARP and Gram-schmidt is both algorithms are the best pan sharpening algorithms I currently know. These algorithms can preserve either spectral integrity from multispectral bands or spatial details from panchromatic band. The picture below is the comparison result that visualized on ERDAS IMAGINE GLT viewer.

images viewed in GLT viewer



From the picture, we can see that Ehlers fusion giving superior spatial details than Gram-schmidt (GS) and PANSHARP. GS and PANSHARP still leave little artifacts (spectral residuals from original multispectral dataset, especially in bright objects) on the sharpened imagery whereas in the ehlers fusion result, they re totally blended.


I have tried to assess the fusion results using correlation analysis between original multispectral bands and the sharpened bands. Below are the scatter plot results from band 1 (VIS Blue) comparison.





yun zhang


And below are the correlation analysis results.

band x (original bands) band y (sharpened bands) Linear regression equation r
ori_1 YZ_1 0,96x + 10,37 0,96
ori_2 YZ_2 0,96x + 14,92 0,96
ori_3 YZ_3 0,96x + 10,9 0,96
ori_4 YZ_4 0,96x + 14,16 0,96
ori_1 ehlers_1 0,83x + 84,54 0,83
ori_2 ehlers_2 0,88x + 41,99 0,89
ori_3 ehlers_3 0,93x + 9,29 0,95
ori_4 ehlers_4 0,97x + 11,44 0,97
ori_1 GS_1 0,90x + 23,34 0,94
ori_2 GS_2 0,91x + 32,04 0,94
ori_3 GS_3 0,94x + 15,66 0,95
ori_4 GS_4 0,99x + 4,57 0,96


Explanation : 1 = band 1 (blue); 2 = band 2 (green ); 3 = band 3 (red); 4 = band 4 (NIR)


Correlation analysis results showed that Yun Zhang PANSHARP algorithm give stable spectral preservation at all bands. Opposite with the spatial quality analysis result, Ehlers fusion can’t maintain the spectral preservation at all bands than GS or YZ PANSHARP. Although at band 4 (NIR) Ehlers give the best correlation with the original band, the correlation was dropped on the other bands. Although some algorithm give better result than the others, we can concluded that all the three algorithms give satisfied spectral preservation result and high correlation coefficient with the original multispectral bands. These algorithms are better than the more traditional algorithms like IHS fusion, CN Spectral sharpening, Brovey, and PC fusion.

Comment please.



  1. bayu andrianto says:

    Good article, but i think for compare the quality of pan-sharpened images, correlation analysis is not enough. You must do a blur analysis, because the quality of the pan-sharpened images are not quantify by the spectral correlation. The quality is measured by ability of the result image for displaying object detailed. Blur analysis can measure object details that lost when we do fusion process. I ever build IDL script for doing this process, i will happy if i can share it with you.

    Ps: I think the best pan sharpened (by object detailed that their resulted) in this case is the methods that use signal processing transformation. Using signal processing transformation can keep and simplify the quality of the result image because the don’t use spasial domain that only doing analysis pixel by pixel and only focus in color quality. So i think ehlers is the best, you can check and search again for futher explanation about this problem.


    Bayu Andrianto

    • pixelcooker says:

      thank you brotha, yup, qualitative assessment based on our perception about result image (artifacts identification, color preservation, etc), of course isnt enough, its still subjective and contain some uncertainties if the evaluator are different, blur analysis, i just hear it, okay, please share the script to me,

      thank you again,

  2. yuhendra says:

    i am just to learn, how to calculate correlation analysis

    • pixelcooker says:

      thanks for comment, doing correlation analysis is easy, you can do that task in spreadsheet software like excel or open office calc, what you have to do is just extract pixel values from bands you desired, plot it in 2 or more column table, and perform the correlation, if yu oprefer automatic way, PCI geomatica have module to do that, in this post , I am using PCI geomatica, and oh almost forgot, ILWIS also able to perform this kind of analysis automatically, and its freeware, 😉

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s