[Retracted] A Deep Learning Approach for Recognizing the Cursive Tamil Characters in Palm Leaf Manuscripts

Table 1

Summarization of different parameters of the proposed ThimuNet.

Convolutional layer	Parameters	Neurons	Connections	Activations
Layer 0- INPUT (64 × 64) 3 channels	0	0	0	64 × 64 × 3 A = 12,288
Layer 1- FILTER (5 × 5) 32 outputs	Wt = 32 × (5 × 5) × 3 B = 32 P = (2400 + 32) = 2,432	(64 × 64) × 32 N = 1,31,072	C = 32,76,800 (64 × 64) × 32 × (5 × 5)	Conv = 64 × 64 × 32 pool = 32 × 32 × 32 BN = 64 × 64 × 32 A = 294,912
Layer 2- FILTER (5 × 5) 64 outputs	Wt = 64 × (5 × 5) × 32 B = 64 P = (51200 + 64) = 51,264	(64 × 64) × 64 N = 2,62,144	(64 × 64) × 64 × (5 × 5) C = 65,53,600	Conv = 32 × 32 × 64 BN = 32 × 32 × 64 pool = 16 × 16 × 64 A = 147,456
Layer 3- FILTER (5 × 5) 128 outputs	Wt = 128 × (5 × 5) × 64 B = 128 P = (204800 + 128) = 2,04,928	(64 × 64) × 128 N = 5,24,288	(64 × 64) × 128 × (5 × 5) C = 1,31,07,200	Conv = 16 × 16 × 128 BN = 16 × 16 × 128 pool = 8 × 8 × 128 A = 73,728
Layer 4- FILTER (5 × 5) 256 outputs	Wt = 256 × (5 × 5) × 128 B = 256 P = (819200 + 256) = 8,19,456	(64 × 64) × 256 N = 10,48,576	(64 × 64) × 256 × (5 × 5) C = 2,62,14,400	Conv = 8 × 8 × 256 BN = 8 × 8 × 256 pool = 4 × 4 × 256 A = 36,864
Layer 5- FILTER (5 × 5) 512 outputs	Wt = 512 × (5 × 5) × 256 B = 512 P = (3276800 + 512) = 32,77,312	(64 × 64) × 512 N = 20,97,152	(64 × 64) × 512 × (5 × 5) C = 5,24,28,800	Conv = 4 × 4 × 512 BN = 4 × 4 × 512 pool = 2 × 2 × 512 A = 18,432
Fully connected layer (FC layer)—46 class outputs	Wt = 512 × (2 × 2) × 46 B = 46 P = 94,208	0	0	A = 46