The following figures show the recognition accuracy for each object using the local system only. For each objecr, I used 10 times, each time in a random orientation in a clustered scene with two other objects. The two other objects did not have a similar shape to the target object. The distracles were otherwise randomly selectected. The results are shown below.