Pattern Recognition Letters
Lighting Independent Skin Tone Detection Using Neural Networks
IEICE - Transactions on Information and Systems
| Hi-index | 0.01 |
This paper describes a face detection and recognitionsystem in color image sequences with a novel schemeto model skin color in the RGB color-space usingneural networks. In our approach, there are nolimitations regarding human skin color. This methodeliminates the difficulty of describing non-skin samplesby approximating non-skin color from skin samples inthe VLSI Systems Laboratory skin database. Theneural network algorithm based face detection isperformed by using a multilayer feed-forward neuralnetwork trained with back-propagation learningalgorithm in conjunction with a modular approachutilizing the distance based learning for reducing thestructural complexity of the network by analyzing eachframe in the video sequence. The recognition isperformed based on Composite Principal ComponentAnalysis (CPCA) algorithm. This algorithm is betterequipped to recognize faces under the conditions ofvarying illumination and pose compared to theconventional PCA. The system is capable of detectingand recognizing faces at the rate of 10 frames persecond when the frame resolution is 320 脳 240 and thecolor depth is 24-bit.