It is important to note that approximately 3% of the banana samples, 8% of the irregular samples, and 6% of the good samples were considered borderline samples. They did not have distinct shape characteristics and could be categorized into more than one shape category depending on human inspector’s interpretation of the shape. The proposed TAC similarity measure algorithm categorized approximately half of those borderline samples differently than human. If these borderline oysters were to be excluded from the test samples, the performance accuracies would have been approximately 4% and 3% higher for irregular and good, respectively.