Skin Detection Using Hybrid Colour Space of RGB-H-CMYK

Abstract

Skin detection is an essential step in human face detection and/or recognition, using digital image processing techniques. This paper presents a new human skin detection technique, termed as RGB-H-CMYK that uses triple colour spaces of an image, namely RGB (Red, Blue and Green), H (Hue of HSV) and CMYK (Cyan, Magenta, Yellow and Black). In this proposed method, threshold-based rules are applied on RGB, H and CMYK for skin classification. The input image in these three hybrid colour schemes is explored in different combination such as RC (RGB and CMYK), RH (RGB and H) and RHC (RGB and H and CMYK). The RHC_Vote qualifies the current pixel as skin pixel when at least two rules vote for it. The computational merit of this hybrid colour scheme-based skin detection is validated on the real-time dataset and ECU skin database. The average Recall and Accuracy of this method is recorded as 85% and 89%, respectively. This approach is confirmed to have an edge over its competitive methods, as it promises object localization, based on neighbourhood intensities without using the computationally complex approaches such as facial texture and geometric properties.