Robust Multi-Colour-Based Skin Detection

Automatic skin detection is a component of various imaging applications, such as face detection and tracking, content categorization, image enhancement, adaptive compression, etc. Colour-based methods have proven to be well suited for this task, but generally suffer from a type of false detection which adversely influences the aforementioned tasks, namely the confident detection of hair regions as skin. This paper presents a new fast and robust skin detection method which addresses this problem by relying on a multiple classifier configuration of skin and hair colour models in multiple colour spaces. Robust Multi-Colour-Based Skin Detection S. Paschalakis and M. Bober Visual Information Laboratory Mitsubishi Electric Information Technology Centre Europe B.V. The Surrey Research Park, Guildford, Surrey GU2 7YD, UK e-mail: {stavros.paschalakis, miroslaw.bober}@vil.ite.mee.com Abstract: Automatic skin detection is a component of various imaging applications, such as face detection and tracking, content categorization, image enhancement, adaptive compression, etc. Colour-based methods have proven to be well suited for this task, but generally suffer from a type of false detection which adversely influences the aforementioned tasks, namely the confident detection of hair regions as skin. This paper presents a new fast and robust skin detection method which addresses this problem by relying on a multiple classifier configuration of skin and hair colour models in multiple colour spaces. Automatic skin detection is a component of various imaging applications, such as face detection and tracking, content categorization, image enhancement, adaptive compression, etc. Colour-based methods have proven to be well suited for this task, but generally suffer from a type of false detection which adversely influences the aforementioned tasks, namely the confident detection of hair regions as skin. This paper presents a new fast and robust skin detection method which addresses this problem by relying on a multiple classifier configuration of skin and hair colour models in multiple colour spaces.