A New Dataset Size Reduction Approach for PCA-Based Classification in OCR Application

Table 2

Result comparison for handwritten Farsi/Arabic digit recognition.

Year	References	Training dataset	Total number of testing samples	Number of features	Classifier	Accuracy
2003	Mowlaei and Faez [27]	(Private) (only 8 digits)	1600	64	SVM	92.44%
2003	Sadri et al. [28]	CENPARMI ver. 1	3035	64	MLP-NN	91.25%
2004	Mozaffari et al. [29]	(Private) (only 8 digits)	1600	64	MLP-NN	91.37%
2005	Mozaffari et al. [25]	(Private) (only 8 digits)	1600	240	Fr. NN	86.30%
2007	Ziaratban et al. [31]	(Private)	4000	60	MLP-NN	97.65%
2011	Enayatifar and Alirezanejad [32]	Hoda (60,000 samples)	20,000	48	MLP-NN	92.70%
2013	The proposed method	Hoda (60,000 samples)	20,000	79	𝑘-NN	97.11%
2013	The proposed method	Hoda (30,000 samples)	20,000	79	𝑘-NN	96.39%
2013	The proposed method	Hoda (20,000 samples)	20,000	79	𝑘-NN	94.08%
2013	The proposed method	Hoda (15,000 samples)	20,000	79	𝑘-NN	93.57%
2013	The proposed method	Hoda (60,000 samples)	20,000	79	MLP-NN	96.27%
2013	The proposed method	Hoda (30,000 samples)	20,000	79	MLP-NN	95.14%
2013	The proposed method	Hoda (20,000 samples)	20,000	79	MLP-NN	90.71%
2013	The proposed method	Hoda (10,000 samples)	20,000	79	MLP-NN	82.86%