Novel framework for segmentation of skin regions using chromatic and textural information

No Thumbnail Available
Journal Title
Journal ISSN
Volume Title
Skin detection is an important step in various image processing and vision-based Human-Computer Interaction (HCI) applications. It is the process of finding skin-coloured pixels and regions in an image or a video. The major challenges of skin detection in images are -- presence of skin-like colours in background and changes in chromatic appearance of skin regions due to non-uniform illumination. In addition to these problems, detection of skin regions in videos is more challenging in presence of time-varying illumination conditions and dynamic backgrounds. Motivated by these facts, we have proposed a set of skin detection algorithms for different environmental conditions using chromatic and textural properties of skin regions. A new probability map termed as discriminative space map (DSM) is proposed by extracting most discriminative features between skin and non-skin regions. A novel adaptive discriminative analysis (ADA) is proposed to extract most discriminant features between skin and non-skin regions from an image itself in an unsupervised manner. Subsequently, a dynamic region growing (DRG) method is employed to allow skin regions to grow dynamically. To handle effect of non-uniform illumination on skin colour, a novel skin detection method is proposed by utilising an image pixel distribution model (IDM), which is derived using a Gaussian Mixture Model (GMM) in a given colour space. In this method, a local skin distribution model (LSDM) and a local background distribution model (LBDM) are derived by exploiting the similarity between the IDM and a reference skin pixel distribution model. The reference skin model is derived from a set of facial skin pixels, and it is termed as facial skin distribution model (FSDM). A local skin probability map (LSPM) can be derived using the LSDM and the LBDM.
Supervisor: M.K. Bhuyan