Robust Visual Segmentation using RCrR Plane and Mahalanobis Distance
Share
Date
2014-12-20
Director(s)
Publisher
Universidad Antonio Nariño
Campus
Faculty
Program
Degree obtained
Document type
COAR type
http://purl.org/coar/resource_type/c_6501
Citation
Bibliographic Managers
item.page.resume
In this paper a robust algorithm against illumination changes for skin detection in images is proposed. A database with 50 controlled condition images and 50 without controlled conditions of people in frontal position showing face, hands and arms was used. Five algorithms to perform color correction are evaluated: Simple Correction with Green Channel, Color Channel Compression, Color Channel Expansion, Fixed Reference and Gamma Correction. And four algorithms for segmentation are evaluated as well: RGB Skin Color, Reference Histogram, Euclidean Distance and Mahalanobis Distance. The proposed algorithm uses the Fixed Reference method together with Gamma Correction for color correction and performs the skin segmentation based on an RCrR color plane, found by making the transformation of the images using RGB and YCbCr spaces, finally Mahalanobis Distance is used. An average sensitivity value of 99.36 % and specificity of 84.31 % were obtained as result.