Cross-Modal Discrimination Hashing Retrieval Using Variable Length
Algorithm 1
Training produce of proposed method.
Input: Training datasets and label matrix ; Initialized association matrix ; Initialized variable-length hash ; Initialized iteration control parameter
Output: Variables
Procedure:
(0)
Applying label matrix and (9) to construct a semantic similarity matrix
(1)
;
(2)
while do
(3)
According to (14), update the dictionary projection matrix ;
(4)
According to (16), update the association matrix ;
(5)
According to equation (18) and the detailed solving process in Ref. [14], the hash code of variable length is updated one line at a time and finally updated as a whole ;
(6)
If the objective function (12) tends to converge, and stop the iteration; otherwise, skip to step (2);