Robust rotation invariant multi-view eye localization method
一种鲁棒的旋转不变性多姿态人眼定位方法*

Focused on eyes localization on multi-view face with arbitrary rotation in plane, this paper proposed a new and robust method. First, located the face by the rotation invariant multi-view (RIMV) face detector and determined the eye searching range on the face. Then, adopted the crossing detection method to scan the searching regions, which used the brow-eye feature and the improved two-level SVM detectors trained by large scale multi-view brow-eye and eye examples. Selected the window ones with the highest values in the SVM discriminant function as the candidate eye regions. Finally, after filtering and merging the candidate eye regions based on their overlapping states, located the exact eyes. Experiments show that the method has high accuracy and strong robustness to the eye localization with arbitrary face pose and expression in the complex background.