Toward Face Detection, Pose Estimation and Human Recognition from Hyperspectral Imagery.

Yi-Ting Chou and Peter Bajcsy

Technical Report NCSA-ALG04-0005, October 2004

We present our work on face detection and pose estimation from hyperspectral imagery. The long-term goal of our research is to explore the use of hyperspectral imagery for building an automated system for human recognition. We report our preliminary results obtained with the face detection algorithm proposed by Paul Viola and Michael Jones. The algorithm was extended to work not only with monochromatic images but also with any dimensional images by pre-processing images using principal component analysis (PCA) and applying the algorithm to the first principal component.

The second part of this report focuses on spectral-based and spatial-based pose estimation. Due to the complexity of human head pose estimation; we analyze the problem of human hand pose estimation in our initial study. We investigate the hand material properties by spectral and spatial analyses, and identify the primary limitations of each analysis from the pose estimation viewpoint. A summary of pros and cons for spectral and spatial analyses is provided at the end.