Urdu Handwritten Characters Data Visualization and Recognition Using Distributed Stochastic Neighborhood Embedding and Deep Network

Table 5

Comparison Analysis of our proposed approach for Urdu handwritten character classification with state-of-the-art approaches.

Reference	Approach	Features	Accuracy (%)
Accuracy rate of classification of the Urdu handwritten character
Ali et al. [28]	Neural network	Geometrical strokes	75–80%
Haider and Khan [29]	BPNN, PNN	Geometrical strokes	66%
Shahzad et al. [30]	Linear classifier	Statistical features	66%
Ahmed et al. [31]	BLSTM	Pixel-based	92–94%
Ko and Poruran [32]	SVM	Transfer-learning features	82.30%
Ali et al. [33]	SVM	Pixel-based features	95.79%
Our approach	CNN	Pixel- and geometrical-based	96.04%
Accuracy rate of classification of Urdu handwritten numeral
Borse and Ansari [34]	Daubechies wavelet	Pixel-based	92.05%
Razzak et al. [35]; Razzak et al. [36]	HMM, fuzzy rule	Pixel-based	97.45%, 97.09%
Sarkhel et al. [37]	Multi-column multi-scale CNN	Non-explicit feature based approach	98.90%
Takruri et al. [38]	(MMCNN) fuzzy C-Means classifier ANN	Pixel-based features	88.00%
Said et al. [39]	ANN	Pixel-based features	94.00%
Mowlaei et al. [40]	ANN	Wavelet-based features	97.34%
Our approach	CNN	Pixel- and geometrical-based	99.01%