Hand posture recognition using the intrinsic dimension

This paper presents a hand posture recognition algorithm based on the intrinsic dimension (ID). Hand segmentation is reliable with cluttered background due to the use of the time-of-flight (ToF) range camera. The hand pose recognition is performed using an image indexing technique with a reference database of 360 images. The database and the query images consist of nine types of hand postures. The similarity measure used for finding the best match with the query image is the correlation between the 1D representations of the 2D ID hand contours. The average recognition rate is above 96%, which is satisfactory.