Summer internship at SAC, ISRO, Ahmedabad in 2021
https://ieeexplore.ieee.org/document/9791967
An Improved IHS Image Fusion Algorithm using Medoid Intensity Match and Bilateral Filter
I interned in the Signal and Image Processing Group at Space Applications Centre, ISRO, Ahmedabad. It was a lovely experience working with the scientists.
About Space Applications Centre, ISRO
Space Applications Centre (SAC) is a premier research institution located in Ahmedabad, India. It is a part of the Indian Space Research Organization. It is engaged in research, development and demonstrations of applications in various fields such as telecommunications, remote sensing, meteorology and satellite navigation.
Background of my work
Image fusion is a productive method to use a large capacity of data. This blog contains remote sensing satellite images and its data analysis, the basic image fusion methods which I have carried out. This blog also contains the novel image fusion method developed with the experimental results of the following methods along with the quantitative assessment which allows us to get a clear idea in the back of our mind which of the following methods are the most successful. Our novel IHS improved method comes out to be the most effective way of all the basic fusion methods.
My Work
I studied and worked upon all the traditional image fusion methods which are: Brovey Transform, Intensity Hue Saturation(IHS), and Principal Component Analysis(PCA).
The Brovey Transform utilizes a Multispectral (Resource sat) band and Panchromatic (Carto sat) band. Every multispectral band is multiplied by a proportion of the panchromatic band divided by the amount of the MS bands. The spatial data is very much protected in this technique yet this strategy leads to spectral distortion in the outcomes.
IHS is the most typical image fusion method in which the image experiences feature enhancement with the improvement of spatial resolution. The R, G, and B bands of MS images are converted to IHS components, replacing the intensity with the spatial information whereas the hue and saturation components with spectral information. IHS enhances the spatial quality of the MS image but as result, it leads to spectral distortion.
PCA transformation is a reduction method of dimensions of multivariate data of images which transforms enormous data sets into various smaller data sets while protecting as lots of information as possible. This is followed by a three-step process: standardization of the dataset, computation of covariance from the matrix, and finding principal components by computing eigenvalues and eigenvectors. The main advantage is a large number of data inputs can be compressed into a much smaller amount of outputs whereas the disadvantage is after the computation of the eigenvector, only the first value is selected for 90% of the shared information, which in turn leads to the loss of information.
Thus, I developed a novel image fusion technique with my mentor which is termed “Improved IHS Image fusion”. Medoid Intensity Match and Bilaterally filtering of multispectral and panchromatic images have been done for the novel improved IHS image fusion method which comes out to be the best of all the traditional image fusion methods. Medoid Intensity Match was carried out on the multispectral image i.e. Resourcesat-2 Indian remote sensing satellite image and bilateral filtering on the panchromatic image i.e. Cartosat-2 satellite image.
Results
The proposed improved IHS image fusion technique is evaluated with Cartosat-1 panchromatic data and Resourcesat-2 multispectral data. The data is selected such that multi-sensor data have the closest date of imaging acquisitions. The image fusion methods are assessed using a different metric that includes Structural Similarity (SSIM), Correlation Coefficient (CC), Root Mean Square Error (RMSE), Spectral Angle Mapper (SAM) ], Erreur Relative Globale Adimensionnelle de Synthe `se (ERGAS) and Universal Image Quality Index (UIQI). Bold indicates the best value of each metric among different image fusion methods. It has been observed that Improved IHS image fusion performed better in all the indices except SSIM. The visual assessment also confirms that improved IHS has better spectral and radiometric characteristics while blending the spatial details from the Cartosat image.
Conclusion
The improved IHS through medoid intensity match from multispectral image and bilateral filtering of the panchromatic image has led to the improvement of the fused image. This process is better than the other component substitution-based image fusion techniques in terms of both spatial and spectral quality. It can also be observed visually in the experiment analysis section and quantitatively using different image fusion metrics.
Guidance From Bennett University
Bennett University was very helpful in providing me with a Letter of Recommendation to work in such a prestigious organization. The courses which were taught helped me a lot. Professors have been amazing throughout and helped me at every moment whenever required.
