Modified Multi-Level Pattern Histograms (MLPH) for SAR image classification.
Some results for different regions in and around New York City area:
Platform tested on: Ubuntu 16.04
Open the terminal inside the directory “IP_Project_Group23”
To run MLPH on training images and train SVM & test on hold-out set:
$ python -m sarclf -run_mlph 1 -train_svm 1
$ python -m sarclf
$ python -m sarclf -test 0 -clfimg <image_path>
$ python -m sarclf -test 0 -clfimg ./test_images/2.tif
$ python -m sarclf -test 0 -clfimg ./test_images/2.tif -modified 1
Following the research on MLPH, we derive a pattern matrix for each pixel based on a threshold value, use these matrices varying bin lengths to get local pattern sub-histograms, concatenated to give local pattern histogram. With multiple thresholds, the concatenation of local pattern histograms gives MLPH for each pixel.