Maximum Noise Fraction Toolbox
Features:
Identifies the noise-removed principal components from the hyperspectral data.
Combines these principal components in a pseudo-color image.
Plots spectra of components.
Quantifies the area by each component.
Saves MNF-PCA data in IDCube format.
Steps:
1. Select Maximum Noise Fraction from the Toolbox tabs from the Main Interface.
2. In the COMPONENT SELECTION panel scroll through different principal components assigned to different color channels. (Default Component #1 is red, component #2 is green, and component #3 is blue). Note: Only the first 20 principal components can be accessed.
3. Assign Weight to any of the channels to enhance the signal at a specific channel. Default values =1 for each channel.
4. Image enhancement enables improved visualization. Current options are:
a. Contrast Saturation Limit (%). The default value is 0. The range is from 0 to 49%.
b. Gamma Correction. Gamma Correction takes values between 0 and infinity. The default value is 1. If gamma is <1, the image is weighted toward brighter pixels, if gamma is >1, the image is weighted toward darker pixels.
5. Click Quantify to open a dialog with the following options.
6. (Optional) Set Threshold value, %. The default level is 30% and will be applied to each component. You do not need to change the value at this moment; you can do that in the next step.
7. Select the Principal Components number(s) that you would like to quantify. Use the format such as 1 2 3 4 with spaces between the numbers. NOTE: The limit is 10 components.
8. Select an Overlay method. The available options are:
a. SAM (Spectral Angle Map) – default.
b. SID (Spectral Information Divergence).
c. SIDSAM (Spectral Information Divergence-Spectral Angle Mapper Hybrid Method).
9. Click Apply Overlay. The new pop-up NMF Quantification will show the following information:
a. RGB rendering of the original dataset.
b. Threshold values, % selection box for individual Principal Component (the limit is 10 components).
c. Thresholded image (based on the input of the threshold value) with the overlaid principal components.
d. Histograms that correspond to each of the components.
e. Area of each component in pixels (located in the title of each histogram plot component).
Adjust the Threshold value for each component. Note that the colors might overlay with the map of the early listed principal component number being on top. To verify overlap, assign the early component Threshold value to 0.
Tips:
Each image can be zoomed and panned by using a mouse. Each image can be also modified using shortcuts.
You can click on each image to see it in a larger separate window. Each image can be saved.
Additional Information:
MNF toolbox computes the specified number of principal component bands by using the Maximum Noise Fraction (MNF) transform. To achieve spectral dimensionality reduction, the specified number of principal components must be less than the number of spectral bands in the input datacube. MNF computes the principal components that maximize the signal-noise-ratio, rather than the variance. MNF transform is particularly efficient at deriving principal components from noisy band images. The principal component bands are spectrally distinct bands with low interband correlation.
The components derived using MNF transform are also called non-adjusted principal components and the MNF transform arranges principal components in the decreasing order of the principal component’s image quality.
MNF transform is equivalent to a two-stage transformation in which the data are first transformed so that the noise covariance matrix is the identity matrix and the second stage is the principal components transform.
MNF transformation gives better signal-to-noise ratio (SNR) than PCA transformation when signal-dependent noise is present in the input data. PCA transformation works better than MNF transformation when Gaussian white noise is present in the input data.
The toolbox includes the following set of algorithms: a) MNF (noise removal +PCA) and b) Overlay which includes three algorithms: SAM, SID, and SAMSID. The calculations are similar to the abundance maps method used in the endmembers mapping but using principal components instead of endmembers.
References:
MNF: Green, A.A., M. Berman, P. Switzer, and M.D. Craig. “A Transformation for Ordering Multispectral Data in Terms of Image Quality with Implications for Noise Removal.” IEEE Transactions on Geoscience and Remote Sensing 26, no. 1 (January 1988): 65–74. https://doi.org/10.1109/36.3001.
SAM: Kruse, F. A., A. B. Lefkoff, J. B. Boardman, K. B. Heidebrecht, A. T. Shapiro, P. J. Barloon, and A. F. H. Goetz. "The Spectral Image Processing System (SIPS) - Interactive Visualization and Analysis of Imaging spectrometer Data." Remote Sensing of Environment 44 (1993): 145-163.
SID: Chein-I Chang. “An Information-Theoretic Approach to Spectral Variability, Similarity, and Discrimination for Hyperspectral Image Analysis.” IEEE Transactions on Information Theory 46, no. 5 (August 2000): 1927–32. https://doi.org/10.1109/18.857802.
SIMSAD: Chang, Chein-I. “New Hyperspectral Discrimination Measure for Spectral Characterization.” Optical Engineering 43, no. 8 (August 1, 2004): 1777. https://doi.org/10.1117/1.1766301.
