Enhancing the Power of CNN Using Data Augmentation Techniques for Odia Handwritten Character Recognition

Table 13

A comparative study for multilingual handwritten character and numeral recognition.

Reference	Language (dataset)	Methods	Digit recognition accuracy	Character recognition accuracy
[24]	CMATERdb-Bangla	DenseNet	99.13	98.31
[25]	CMATERdb-Bangla	Modified ResNet-18	—	95.10
[23]	Self-prepared database-Bangla	Celled projection (CP) + k-NN	94.12	—
[22]	CMATERdb-Bangla	LRF, HOG and diagonal feature + SVM	—	88.73
[45]	DHCD-Devanagari	Deep CNN		98.47
[46]	Self-prepared database-Devanagari	Chain code histogram and moment invariant features + MLP		98.03
[47]	Self prepared database-Devanagari	Curvelet transform and the character geometry + k-NN		93.8
[48]	ISI Kolkata numeral database	CNN with genetic algorithm	96.41
[49]	CMATERdb-Telugu	Discrete wavelet transform (DWT), projection profile (PP) and singular value decomposition (SVD) + k-nearest neighbor (k-NN) and support vector machine (SVM)		95.47 (SVM on DWT features)
[50]	Telugu database	Binary external symmetry axis constellation features + quadratic discriminate classifier and SVM		80.6 (SVM); 87.6 (QDA)
[51]	MNIST, CMATERdb, ISI, Gujurati and Punjabi badabase	CNN	96.23