A Learning-Based Prediction-and-Verification Segmentation Scheme for Hand Sign Image Sequence

This paper presents a prediction-and-veriication segmentation scheme using attention images from multiple xations. The scheme has two major components: (a) a hierarchical quasi-Voronoi diagram which organizes training attention images for prediction of the segmentation masks; (b) a learning-based function approximation scheme to verify the segmentation result. A major advantage of this scheme is that it can handle a large number of diierent deformable objects presented in complex backgrounds. The scheme is also relatively eecient since the segmentation is guided by the past knowledge through a prediction-and-veriication scheme. The system was tested to segment hands in sequences of intensity images, where each sequence represents a hand sign in American Sign Language. The experimental result showed a 95% correct segmentation rate with a 3% false rejection rate.