Skin Segmentation

General Info

Thesis by: Alexander Schoch
Advisor: Professor Nassir Navab
Supervision by: Alexandru Duliu
Contact: Alexandru Duliu
Status: Finished

Motivation

Skin segmentation plays an important role in multi-view reconstruction of human limbs from color or multispectral images. In this approach regions of the input images are segmented in order to produce an initial estimate of the regions where skin is present. These regions will be further processed in later stages to extract a mask of the object of interest. These masks play an important role in multi-view reconstructions as they offer information about which pixels are relevant for the reconstruction and also provide the silhouette of the object, increasing the accuracy and reliability of the reconstruction.

Goals

The project will be divided into multiple work-packages:

• Colorimetric Segmentation involves that the student implement a robust detection of foreground and background pixels. This stage will also assume that image pixels geometrically outside the background be excluded from the segmentation.
• Random Walk Segmentation will require reliable seed pixels on both foreground and background. The algorithm will then effectively separate the two regions from each other. The student will be tasked with integrating a C++ implementation of this algorithm.

References

Author	Title	Year	Journal/Proceedings	Reftype	DOI/URL
Gasparini, F. & Schettini, R.	[1] Skin segmentation using multiple thresholding	2006	Vol. 6061Proceedings of SPIE, pp. 128-135	inproceedings	URL
Phung, S.L., Bouzerdoum Sr, A. & Chai Sr, D.	[2] Skin segmentation using color pixel classification: analysis and comparison	2005	Pattern Analysis and Machine Intelligence, IEEE Transactions on Vol. 27(1), pp. 148-154	article	URL
Phung, S.L., Chai, D. & Bouzerdoum, A.	[3] Adaptive skin segmentation in color images	2003	Vol. 3Multimedia and Expo, 2003. ICME'03. Proceedings. 2003 International Conference on, pp. III-173	inproceedings	URL