Triplet Deep Hashing with Joint Supervised Loss Based on Deep Neural Networks

Algorithm 1

The procedure of JLTDH.

	Input: training images ; code length ; epochs = 150; superparameters ; and minibatch = 128
Initialization: initialize neural network parameters , , and with the AlexNet model and iteration number
Generate triplet training set: ;
for ; ; do
Randomly sample a minibatch of points from , and for each sampled point :
(i)	Calculate by forward propagation;
(ii)	Compute ;
(iii)	Compute the binary code of with ;
(iv)	Compute derivatives for point ;
(v)	Update the parameters by utilizing BP;
(vi)	Compute ;
for ; ; do
Discrete cyclic coordinate descent (DCC) optimization:
(i)	Compute according to (16);
(ii)	Iteratively optimization update bit by bit using the DCC method according to (21) in the minibatch;
End for
Update in the minibatch according to joint loss function in (13);
End for
Output: , with the parameters ;
Hash codes .