Table 1.

Comparison of the proposed method with previous methods.

Category	Methods	Strengths	Weakness
Non-pre-classification-based method	-Using GLVQ [10]	Additional processing time for pre-classification is not required	Accuracy enhancement is limited because of the large number of classes of banknotes, including two sides (obverse and reverse) and two directions (forward and backward)
	-Using SURF features [11]
	-Using QWT, generalized Gaussian distribution, and BP neural network [12]
	-Using local PCA, SOM, and LVQ [13,14,15,16,17]
	-Using correlation measure [19,20]
	-Using SURF and the spatial relationship of matched SURF features [21]
Pre-classification-based method	-Using BP neural network [9]	The number of classes of banknotes can be reduced four-fold, because of the pre-classification of the two sides and two directions	The classification accuracy of banknote type is not presented
	-Using SVM classifier with PCA features (proposed method)		Additional processing time is required for pre-classification