Sample results of the Multiple Color Image Fusing Method


Here are some results of the Multiple Color Image Fusing Method.

Result 1

This test data was artificially created by dividing an original image into three fragments, top, left and right.  The color balance and brightness of the left and right images were changed with Microsoft Photo Editor.  The left image (left middle) is reddish.  The right image (left bottom) is greenish.  The images below show the overlapping region images of each fragment (top, left and right fragments from the top row).  The top fragment (top row) color space was the target color space, then the left (middle row) and right (bottom row) color spaces were transformed to the top fragment (top row) color space.

Before the Multiple Color Image Fusion
 After the Multiple Color Image Fusion
Top Before
Top (Target Color Space)

Left Before without Noise
Left (Before)

Right Before witout Noise
Right (Before)

Whole Before without Noise
Whole (Before)
 Top After
Top (Target Color Space)

Left After without Noise
Left (After)

Right After without Noise
Right (After)

Whole After without Noise
Whole (After)

In addition, RGB color difference histograms before and after the multiple color image fusion are shown below.  The black dotted line in the histograms shows RGB color differences before the multiple color image fusion.  The blue solid and red dashed lines in the histograms show RGB color differences after the multiple color image fusion and the pairwise color image fusion (initial color image fusion), respectively.  The top row histograms show RGB color differences between the left and top fragments.  The middle row histograms show RGB color differences between the right and top fragments.  The bottom row histograms show RGB color differences between the left and right fragments.  As you can see in the histograms, all peaks have moved to approximately 0 (no error point).  Moreover, RGB color differences between the left and right fragments with the multiple color image fusion is smaller than with the pairwise color image fusion, though RGB color differences between the top and left fragments and between the top and right fragments with both methods are almost same.

Histogram without Noise

Result 2

This test data was created in the same way as above, except Gaussian noise was added to the left and right fragments (the amount of the noise is different for each fragment) with Adobe Photoshop.  The images show the overlapping region images of each fragment (left, right and top fragments from the top row).  The top fragment (top row) color space was the target color space, then the left (middle row) and right (bottom row) color spaces were transformed to the top (top row) fragment color space.

Before the Multiple Color Image Fusion
 After the Multiple Color Image Fusion
Top Before
Top (Target Color Space)

Left Before with Noise
Left (Before)

Right Before with Noise
Right (Before)

Whole Before with Noise
Whole (Before)
 Top After
Top (Target Color Space)

Left After with Noise
Left (After)

Right After with Noise
Right (After)

Whole After with Noise
Whole (After)

In addition, RGB color difference histograms before and after the multiple color image fusion are shown below.  The black dotted line in the histograms shows RGB color differences before the multiple color image fusion.  The blue solid and red dashed lines in the histograms show RGB color differences after the multiple color image fusion and the pairwise color image fusion (initial color image fusion), respectively.  The top row histograms show RGB color differences between the left and top fragments.  The middle row histograms show RGB color differences between the right and top fragments.  The bottom row histograms show RGB color differences between the left and right fragments.  A similar result to that of Result 1 was obtained in this experiment.  Note here the histogram axes are much broader due to noise.

Histogram with Noise

Result 3

This test data was obtained with an off-the-shelf digital camera from different views, left, front and right directions (3 images).  The following figures show a VRML model texture mapped with the three images.  As the black squares show in the left figure, color discrepancies at image boundaries exist.  The white circle in the left figure shows the overlapping region of the three images.  As you can see in the result, the color discrepancies have been removed after the multiple color image fusion. However, the red color region in the right side (the circle in the right image) was a bit lighter.

Before the Multiple Color Image Fusion After the Multiple Color Image Fusion
Balcony Before
 Balcony After

Result 4

Here is another VRML model textured with 30 images.  All surfaces of the building are texture mapped with the images.  As you can see in the result, the color discrepancies have been removed after the multiple color image fusion and the appearance of all surfaces is much more consistent than the original.
Before the Multiple Color Image Fusion After the Multiple Color Image Fusion
whole model before color correction
 whole model after color correction


Result 5

Here is another VRML model textured with 8 images obtained with an off-the-shelf digital camera.  All sides of the monument are texture mapped with the images.  There are some color discrepancies at surface boundaries.  All object faces are seen in 3 images. Face 1 is seen in images 1, 2 and 3. Similarily face 2 is in images 3, 4, and 5, face 3 is in images 5, 6 and 7 and face 4 is in images 7, 8 and 1. The target color space is in the front face of the tower (in image 1) and all image color spaces were transformed to the front face's color space by using the multiple and pairwise image color corrections to compare effects of the color corrections.
Before the Color Image Fusions After the Multiple Color Image Fusion After the Pairwise Color Image Fusion
Clock Tower 1 Before
Image 1 (Target Color Space)

Clock Tower 2 Before
Image 2 (Before)

Clock Tower 3 Before
Image 3 (Before)

Clock Tower 4 Before
Image 4 (Before)

Clock Tower 5 Before
Image 5 (Before)

Clock Tower 6 Before
Image 6 (Before)

Clock Tower 7 Before
Image 7 (Before)

Clock Tower 8 Before
Image 8 (Before)
 Clock Tower 1 After
  Image 1 (Target Color Space)

Clock Tower 2 After
Image 2 (After Multiple)

Clock Tower 3 After
Image 3 (After Multiple)

Clock Tower 4 After
Image 4 (After Multiple)

Clock Tower 5 After
Image 5 (After Multiple)

Clock Tower 6 After
Image 6 (After Multiple)

Clock Tower 7 After
Image 7 (After Multiple)

Clock Tower 8 After
Image 8 (After Multiple)
 Clock Tower 1 After Pairwise
Image 1 (Target Color Space)

Clock Tower 2 After Pairwise
Image 2 (After Pairwise)

Clock Tower 3 After Pairwise
Image 3 (After Pairwise)

Clock Tower 4 After Pairwise
Image 4 (After Pairwise)

Clock Tower 5 After Pairwise
Image 5 (After Pairwise)

Clock Tower 6 After Pairwise
Image 6 (After Pairwise)

Clock Tower 7 After Pairwise
Image 7 (After Pairwise)

Clock Tower 8 After Pairwise
Image 8 (After Pairwise)

The following figure shows RGB color difference histograms between the front face of the tower and face 4 after the color correctons.  The red dashed and blue solid lines show the pairwise and multiple color corrections, respectively.

Histogram Comparison

As you can see the above figures, color discrepancies at image boundaries were almost removed after the color corrections. However, the pairwise image color correction gradually increased errors and there is a large dissimilarity. On the other hand, the multiple image color correction prevented the error from accumulating and there is only a small dissimilarity.

Return to HOME