A Mean Eigenwindow Method for Partially Occluded/Destroyed Objects Recognition

This paper describes a method for recognizing partially occluded and/or destroyed objects using an eigenspace method referred to as a ‘mean eigenwindow’ method that stores multiple partially occluded/destroyed objects in an eigenspace. We have proposed to store similar poses, that may include disturbed shapes, of an object in a particular window referred to as the ‘eigen window’ and, finally, mean of appearances of each window is taken into consideration in order to obtain a generalized eigen window called the ‘mean eigenwindow’. This mean eigenwindow is further used for recognizing an unfamiliar pose, including partially occluded or destroyed shapes, and the object type itself. We have applied the proposed approach to various image situations and the method has successfully performed recognition of an object with up to 20% of occlusion and/or destruction.