The current method for performing this operation (without CyberLand) is thus:

• Manually or semi-automatically, select control points in the images
  
• Match the images using a polynomial or triangular net (necessarily having to resort to a "big surface element")
  
• Fuse the matched images
  

Such a method can obtain reasonable results in flat areas, but in mountainous areas, especially when using oblique photography, it can't completely eliminate the geometric distortion of polynomial rectification or a big surface element relative rectification.

The geometric distortion of imagery from various sensors is different, so a good result is very often not possible. 

The distortion of a SPOT image and a TM image is depicted in Figures 1 and 2. 

Figure 1 SPOT and TM images

Figure 2 Different distortions apparent in SPOT and TM images

In Figure 1, the two images are zoomed to the same size. The SPOT image is annotated A, B, C, D, E, F & G and the corresponding annotation in the TM image is marked A', B', C', D', E', F' & G'. In Figure 2, the upper sections marked AB, BC, CD, DE, EF & FG correspond to the lower sections, marked A'B', B'C', C'D', D'E', E'F' & F'G'. In this situation, partial distortion is very obvious and the match result from a polynomial or triangular net (necessarily requiring processing using a "big surface element") will be very poor, as is shown in Figures 3 and 4. 

 

Figure 3 TM cubic polynomial rectification

Figure 4 TM "big surface element" differential rectification

That was the traditional approach.

The CyberLand Solution

In CyberLand, the above problem is solved using Supresoft's automatic feature extraction and global image matching with relaxation routines to obtain a dense homologous point pair network. 

By making each of these thousands of homologous points a control point, a dense triangular network can be readily built up. The system can then perform a rigorous differential rectification using those "small surface element" triangles, realizing precise image matching. 

The result of the match is thus:

 

Figure 5 TM "small surface element" differential rectification

A comparison of the final fused images: 

Figure 6 Results of matching and fusion with TM bands 3, 4 & 7
a) "Big surface element" matching and fusion	b) "Small surface element" matching and fusion