This is the repository for Google Summer of Code 2012 project Spectral Algorithm Development for Opticks. For more information about Opticks visit http://www.opticks.org.
Opticks uses Spectral Processing Extension to perform hyperspectral and multispectral image analysis. Spectral imaging can allow extraction of additional information the human eye fails to capture with its receptors for red, green and blue. This help in uniquely identifying objects based on their spectral signatures, correctly clustering a image into components and detecting changes in the image. These techniques form an integral part of spectral image analysis.
The additional information present in spectral images can be used to train supervised machine learning algorithms in order to achieve very high accuracy in object identification in an image. Same goes for the unsupervised algorithms used to cluster and detect changes in an image.
The repository contain the following Machine Learning techniques/algorithms:
* ISODATA Clustering
* Change Detection using an EM based approach
* SVM Classification
* Back Propagation Neural Networks
The Plugins:
ISODATA
ISODATA is used to cluster an image. The ISODATA algorithm is similar to the k-means algorithm with the distinct difference that the ISODATA algorithm allows for different number of clusters while the k-means assumes that the number of clusters is known a priori. Here we use ISODATA for clustering spectral image. All the original images below are taken from the Opticks Sample Data. The Code is available here: ISODATA
Original Image
ISODATA running on the image.
After first iteration of ISODATA.
After second iteration of ISODATA.
After third iteration of ISODATA.
After fourth iteration of ISODATA.
The clustered image obtained after the final iteration.
Change Detection using EM
This plugin uses an EM based approach for unsupervised change detection. It is based on Automatic Analysis of the Difference Image for Unsupervised Change Detection.
Image of Delaware Bay taken on 23 September 1999
Image of Delaware Bay taken on 10 July 2001
The algorithm running on difference image
Detected Changes(shown in Black)
Support Vector Machine
SVM Plugin uses SMO algorithm for training. One-Against-All strategy is used for multiclass classification. Linear or RBF kernel can be used for training SVM.
Dialog box for Input parameters
SVM training on input data
Predicting pixels using the trained model
