research

related publications

TWO-LEVEL POSE ESTIMATION USING MAJORITY VOTING OF GABOR WAVELETS AND BUNCH GRAPH ANALYSIS

A two-level approach for estimating face pose from a single static image is presented. Gabor wavelets are used as the basic features. The objective of the first level is to derive a good estimate of the pose within some uncertainty. The objective of the second level processing is to minimize this uncertainty by analyzing finer structural details captured by the bunch graphs. The first level analysis enables the use of rigid bunch graph. The framework is evaluated with extensive series of experiments. Using only a single level, 90% accuracy (within 30 degrees) was achieved on the complete dataset of 1,395 images. Second level classification was evaluated for three sets of poses with accuracies ranging between 67%-73% without any uncertainty.

 

We present a two-level classification framework for the accurate pose determination, so as to determine the face pointing direction. The two-level approach is based upon the rationale that visual cues characterizing facial pose has unique multi-resolutions partial frequency and structural signatures. The first level of the approach has the objective of deriving pose estimates with some uncertainty. First level output confines the poses in a small range so that rigid bunch graph can be used thereafter. The objective of the second level processing is to minimize this uncertainty by systematically analyzing the finer structural details captured by the bunch graphs. Gabor wavelets are used as the features. In the coarse level, every Gabor wavelet response is classified using the subspace projection. Two different subspaces are used to get the best descriptors, which are PCA and Kernel Discriminant Analysis (KDA). The classification results from different Gabor wavelet are combined by majority voting. The first level localizes the poses up to an NxN (N=3) sub-window around the true poses. In the fine level, the pose estimation refined by using rigid bunch graph matching, which utilizes the geometrical details of the salient facial component.

REFERENCES

For more information, see related publications.