Engineering Commons^™

Exploring Invariant Hybrid Color Image Features For Face Recognition Under Illumination Variation, Jayesh Mohan

LSU Master's Theses

In this thesis, a novel analysis framework is presented in order to automate testing response of an image-feature descriptor algorithm for face recognition under different illumination conditions and white balance calibration over intra- and inter-color space. The experimental results on the OPFD database show that our analysis framework finds the least sensitive channel of a color space for recognizing a face under unknown illumination, unknown white balance, and the both unknown illumination and white balance conditions. The results also show the combination of channels in a color space which are best suited face recognition.