Wojciech Michal Matkowski, Frodo Kin Sun Chan and Adams Wai Kin Kong. A Study on Wrist Identification for Forensic Investigation. Image and Vision Computing, vol. 88, August 2019, pp 96-112. https://doi.org/10.1016/j.imavis.2019.05.005
Paper can be found now on ScienceDirect here
Preprint can be found on arXiv here
To acquire the NTU-Wrist-Image-Database-v1 dataset (which was used in the paper), download and fill in the "Data Release Agreement.pdf" file. Print the agreement and sign on page 2. Scan the signed copy and send back to [email protected], [email protected] or [email protected] with title "Application for NTU-Wrist-Image-Database-v1 dataset". A download link to the corresponding dataset will be send to you once after we receive the signed agreement file.
Examples of original wrist images (note: in folder SEToriginalWristImages all images are already flipped).
Examples of segmented and flipped wrist images (in folder SETsegmentedWristImages).
Examples of segmented, flipped and aligned ROI wrist images (in folder SETsegmentedAlignedWristImages).
- current folder should be code
- you can run 4 demos to see how the pipeline works
- you can run demo for each module
in the paper Section 3.1
- download pre-trained ensemble of decision trees superpixel skin classifiers here
- unpack downloaded classfiers and put them into classifiersTrees folder, then put classifiersTrees folder into segmentation folder
- the results are saved in results folder if (write2File = true)
NOTE: you can also visualize some steps if (plotFig = true)
in the paper Section 3.2
- add CPD2 to path and run cpd_make
- the results are saved in results folder if (saveResults = true)
NOTE: you can also visualize some steps (in the paper see Fig. 8) if (plotFlag = true)
NOTE2: to change between proc2 and proc2/3 use procSelector=true or procSelector=false;\
Featrue extraction module uses DSIFT from VLFeat library. Download and Guide to instal here
If you have any questions about the paper please email me on [email protected]