Here you can find some test scripts to segment a certain part over an image sequence (with an initialization)

A few scripts are present in this repository :

• DoP_histograms : script to evaluate visually (with histograms) the consistency (in value) of DoP images for thresholding
• blob_test : a test script to evaluate the performance of a blob detector for tracking a certain part (does not give godd results)
• featurepoint_matching : Test script of an estimation of homogeneous matrix between 2 images
• regiongrowing : Region growing algorithm for segmentation
• seqtoolbox : some functions useful for the final treatment
• test3d : script to visualize with a 3d plot a gradient estimation
• test_hog : test script to evaluate the performances of HOG to solve the problem
• test_sequence : main test file. Implements the latest solution found.

• To track the region of interest, an algorithm based on mimimum finding is used. It does not use any kind of classifier or feature detection.
• Since the "tracking" does not use a classification system, it is compulsary to initialize on the first image of the sequence.
• As it is, cannot work in real time with full resolution. Although it can be using unsupervised algorithm (such as Super Pixel or Felzenszwalb algorithm) and image decimation.
• The tracking works best with AoP images