Improving the Accuracy of Two-Color Multiview (2CMV) Advanced Geospatial Information (AGI) Products Using Unsupervised Feature Learning and Optical Flow

. 2019 Jun 8;19(11):2605. doi: 10.3390/s19112605

Algorithm 3 Generating change maps

Difference Image:
$X_{d i f} = | R e f e r e n c e - M i s s i o n |$
Training Data: Divide $X_{d i f}$ into $h x h$ non-overlapping blocks.
Dictionary Generation: Use the K-SVD algorithm to generate an overcomplete dictionary.
Create Feature Space:
- -
  Generate $h x h$ blocks for each pixel in $X_{d i f}$ where the pixel is in the center of the block.
- -
  Use OMP algorithm to generate the projections of the data onto the dictionary.
Clustering: Use the k-means algorithm to classify the feature space into two classes, e.g. change and no-change.
Change maps: Use the two classes to generate the combined change map. Divide the combined change map into two separate change maps based on the changes that occur in the images.