Invariant object recognition by shape space analysis

This paper describes a new approach to invariant object recognition. In this approach, an object is represented by a set of key points called landmarks. All possible translation, scaling, and rotation the object are placed into an equivalent class and associated to a single point in a complex projective space called the shape space. Object recognition is then achieved by distance calculations in this shape space. This approach is invariant to object translation, scaling, and rotation, and is computationally simple. Our experimental results also indicate that it is insensitive to noise and moderate occlusions.