Open AccessAn Improved Similarity Algorithm Based On Deep Hash and Code Bit Independence
Author(s) -
Jiyao Ding,
Anjun Cheng
Publication year - 2020
Publication title -
iop conference series. earth and environmental science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.179
H-Index - 26
eISSN - 1755-1307
pISSN - 1755-1315
DOI - 10.1088/1755-1315/440/3/032079
Subject(s) - hash function , computer science , code (set theory) , pattern recognition (psychology) , feature (linguistics) , convolutional neural network , algorithm , artificial intelligence , image retrieval , image (mathematics) , theoretical computer science , data mining , linguistics , philosophy , computer security , set (abstract data type) , programming language
Deep convolutional neural networks have been widely used in image retrieval because of their powerful feature representation capabilities. Due to the high efficiency of hash space, many image retrieval algorithms based on deep hashing emerge. Aiming at the high correlation between the codes bits of hash code, this paper proposes a new end-to-end trainable deep hash algorithm to implement the feature grouping so that each code bit can express its unique partial image information. We add slice layers to reduce the repetitive expression of information and add the pairing loss to expand the difference between different categories of images and improve the recognition ability of the model for different categories. After training, we use the hierarchical search strategy to test the retrieval ability of the classification model. The mean average precision and recall of our algorithm can reach 87.5% and 92% respectively, which achieves great result in the related field and has important guiding significance for the future research of hash algorithm.