Full text: Proceedings; XXI International Congress for Photogrammetry and Remote Sensing (Part B7-3)

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
4) Rendering by the fragment shader. 
5) Download the texture which stores the rendering result to 
the system memory. 
When 
2 4 . 
— K < H < —71 ‘ 
3 3 
5. THE TRANSFORMED DOMAIN FUSION ON GPU 
This paper mainly analyses IHS transform fusion and DWT 
fusion. 
G=/a+ S-c°s(/f- 2 */3)) 
cos(/r - H) 
R = I(l-S) 
B = 31-R-G 
5.1 The IHS transform fusion on GPU 
There are two kinds of color coordinate systems (or called color 
space) widely used in image processing: ® Color space 
comprised by Red, Green and Blue, which is called RGB space; 
(D IHS model, which is comprised by Intensity, Hue and 
Saturation. The transformation between RGB space and IHS 
space is called IHS transform 131 . The 3 components of IHS 
space which have their independency can be controlled 
separately and describe the color feature exactly. The process of 
IHS is as follow: 
When 
— n<H<2n'- 
3 
S cos(H — 4tt /3) 
cos(5;r / 3 - H) 
G = I(\-S) 
R = 31 - B -G 
(3) 
(4) 
Normalize the original pixel value R 0 •. G 0 > Bo’ 
R=Ro/L, G=G(/L, B=B(JL; 
Where, L is the gray level of image. 
In forward transformation, the values of L H> S are calculated. 
When R=G=B\ S=H= 0, I=R; 
Otherwise: 
I = 
S = 
h = 
^(R + B + G) 
1-min (R,G,B)/I 
R-G/2-B/2 
arccos 
Vi? 2 +G 2 +B 2 -RG-GB-RB 
(1) 
Firstly, according to packing method 2, RGB 3-bands multi- 
spectral image with low resolution is loaded into three textures, 
and then FBO is created and bound with 6 frame buffers, whose 
binding points are LCOLORATTACHMENTOEXT , 
GLCOLORATTACHMENT1 EXT 
GL COLOR ATTACHMENT5 EXT. Each frame buffer is 
corresponding with an output texture. 
Secondly, forward transformation is executed by a shader in one 
circulation. In this paper, shaders are implemented by OpenGL 
Shading Language (GLSL), and RGB 3-bands multi-spectral 
image with low resolution is regarded as input texture. Set the 
number of color buffer area to be 3 by Draw Buffers Extension 
and the buffers are GLCOLORATTACHMENTOEXT, 
GLCOLORATTACHMENT 1 EXT, 
GLCOLORATTACHMENT2 EXT, which are used to store 
I, H, S. In this paper, according to IHS transformation fusion 
model, two shaders are designed to execute the forward and 
inverse transformation. 
When G>=B: H=h and when G<B: H =2 k - h . 
Histogram matching, the grayscale of high resolution 
panchromatic image is stretched to make the average of its 
grayscale and summation of variance identical with the I 
component of IHS space. 
In inverse transformation, the stretched high resolution image is 
regarded as new I component into calculation. The RGB values 
of original space are computed and the inverse transformation is 
as follow: 
When 0 < H < — 7r '■ 
3 
R = I(l + S co&(H) ) 
cos(;r / 3 - H) 
B = 7(1 -S) 
G = 31-R-B 
In the first rendering, forward transformation shader is used and 
destination buffer will obtain values of 3 components I, H, S. It 
is not suitable for the algorithms which need to store the data 
when exporting statistics or measurements to be executed in 
GPU, so histogram matching is handled by CPU. Download the 
buffer data which store I compenent to system memory and 
execute histogram matching with high resolution panchromatic 
image. The data in I are normalized, firstly, we transform the 
data in I to original grayscale. After matching, the stretched 
high resolution image is normalized and regarded as new I 
component to load into texture. At the same time, the buffers 
that store H, S components are treated as source buffer (put the 
texture data and new I component data together to be input 
stream in the second rendering) and 
GLCOLORATTACHMENT3EXT 
GLCOLORATT ACHMENT 4_EXT 
GLCOLORATTACHMENT5EXT are treated as destination 
buffers. In the second rendering, inverse transformation shader 
is used, and the data in destination buffer are downloaded into 
system to finish IHS fusion. The process of calculation is 
described by figure 3.
	        
Waiting...

Note to user

Dear user,

In response to current developments in the web technology used by the Goobi viewer, the software no longer supports your browser.

Please use one of the following browsers to display this page correctly.

Thank you.