Transformation Invariance in Hand Shape Recognition

In hand shape recognition, transformation invariance is key for successful recognition. We propose a system that is invariant to small scale, translation and shape variations. This is achieved by using a-priori knowledge to create a transformation subspace for each hand shape. Transformation subspaces are created by performing principal component analysis (PCA) on images produced using computer animation. A method to increase the efficiency of the system is outlined. This is achieved using a technique of grouping subspaces based on their origin and then organising them into a hierarchical decision tree. We compare the accuracy of this technique with that of the tangent distance technique and display the results