| Why Image Fusion | | | | images. In satellite imagery we can have two types of |
| Multisensor data fusion has become a discipline to | | | | images |
| which more and more general formal solutions to a | | | | Panchromatic images - An image collected in the |
| number of application cases are demanded. Several | | | | broad visual wavelength range but rendered in black |
| situations in image processing simultaneously require | | | | and white. |
| high spatial and high spectral information in a single | | | | Multispectral images - Images optically acquired in more |
| image. This is important in remote sensing. However, | | | | than one spectral or wavelength interval. Each |
| the instruments are not capable of providing such | | | | individual image is usually of the same physical area |
| information either by design or because of | | | | and scale but of a different spectral band. |
| observational constraints. One possible solution for this | | | | The SPOT PAN satellite provides high resolution (10m |
| is data fusion. | | | | pixel) panchromatic data. While the LANDSAT TM |
| Standard Image Fusion Methods | | | | satellite provides low resolution (30m pixel) multispectral |
| Image fusion methods can be broadly classified into | | | | images. Image fusion attempts to merge these images |
| two - spatial domain fusion and transform domain | | | | and produce a single high resolution multispectral image. |
| fusion. | | | | The standard merging methods of image fusion are |
| The fusion methods such as averaging, Brovey | | | | based on Red-Green-Blue (RGB) to |
| method, principal component analysis (PCA) and IHS | | | | Intensity-Hue-Saturation (IHS) transformation. The usual |
| based methods fall under spatial domain approaches. | | | | steps involved in satellite image fusion are as follows: |
| Another important spatial domain fusion method is the | | | | Register the low resolution multispectral images to the |
| high pass filtering based technique. Here the high | | | | same size as the panchromatic image. |
| frequency details are injected into upsampled version | | | | Transform the R,G and B bands of the multispectral |
| of MS images. The disadvantage of spatial domain | | | | image into IHS components. |
| approaches is that they produce spatial distortion in the | | | | Modify the panchromatic image with respect to the |
| fused image. Spectral distortion becomes a negative | | | | multispectral image. This is usually performed by |
| factor while we go for further processing, such as | | | | histogram matching of the panchromatic image with |
| classification problemial distortion can be very well | | | | Intensity component of the multispectral images as |
| handled by transform domain approaches on image | | | | reference. |
| fusion. The multiresolution analysis has become a very | | | | Replace the intensity component by the panchromatic |
| useful tool for analysing remote sensing images. The | | | | image and perform inverse transformation to obtain a |
| discrete wavelet transform has become a very useful | | | | high resolution multispectral image. |
| tool for fusion. Some other fusion methods are also | | | | Medical Image Fusion |
| there, such as Lapacian pyramid based, curvelet | | | | Image fusion has recently become a common term |
| transform based etc. These methods show a better | | | | used within medical diagnostics and treatment. The |
| performance in spatial and spectral quality of the fused | | | | term is used when patient images in different data |
| image compared to other spatial methods of fusion. | | | | formats are fused. These forms can include magnetic |
| The images used in image fusion should already be | | | | resonance image (MRI), computed tomography (CT), |
| registered. Misregistration is a major source of error in | | | | positron emission tomography (PET), and single photon |
| image fusion. Some well-known image fusion methods | | | | emission computed tomography (SPECT). In radiology |
| are: | | | | and radiation oncology, these images serve different |
| High pass filtering technique | | | | purposes. For example, CT images are used more |
| IHS transform based image fusion | | | | often to ascertain differences in tissue density while |
| PCA based image fusion | | | | MRI images are typically used to diagnose brain |
| Wavelet transform image fusionpair-wise spatial | | | | tumors. |
| frequency matching | | | | For accurate diagnoses, radiologists must integrate |
| Applications | | | | information from multiple image formats. Fused, |
| Image Classification | | | | anatomically-consistent images are especially beneficial |
| Aerial and Satellite imaging | | | | in diagnosing and treating cancer. Companies such as |
| Medical imaging | | | | Keosys, MIMvista, IKOE, and BrainLAB have recently |
| Robot vision | | | | created image fusion software to use in conjunction |
| Concealed weapon detection | | | | with radiation treatment planning systems. With the |
| Multi-focus image fusion | | | | advent of these new technologies, radiation oncologists |
| Digital camera application | | | | can take full advantage of intensity modulated radiation |
| Battle field monitoring | | | | therapy (IMRT). Being able to overlay diagnostic |
| Satellite Image Fusion | | | | images onto radiation planning images results in more |
| Several methods are there for merging satellite | | | | accurate IMRT target tumor volumes. |