Semi-supervised cross-modal Hash retrieval method based on class label transfer

Abstract

The invention discloses a semi-supervised cross-modal Hash retrieval method based on class label transmission, which mainly solves the problems that the existing training data does not have enough class label information and the existing semi-supervised multi-modal method cannot effectively utilize the class label information. The implementation scheme comprises: obtaining feature matrixes corresponding to test data and training data; respectively obtaining class label matrixes of the unsupervised training picture set and the unsupervised training text set through class label transmission; constructing an objective function of supervised hashing, iteratively solving to obtain updated picture and text hash code matrixes and projection matrixes respectively, and accordingly solving hash codes of the test picture set and the test text set; and calculating a Hamming distance between the test data hash code and the training data hash code, sorting the Hamming distance from small to large, and taking previous s corresponding training data as a final query result. Supervision information in the multi-modal semi-supervised training sample can be effectively utilized, retrieval precision isimproved, and the method can be applied to information cross retrieval and data storage.

Description

technical field [0001] The invention belongs to the technical field of information retrieval and pattern recognition, and in particular relates to a semi-supervised cross-modal hash retrieval method, which can be applied to information cross retrieval and data storage. Background technique [0002] With the rapid development of information technology such as the Internet and social media in recent years, the data accumulated by all walks of life has shown an explosive growth trend. Moreover, today's data is not only large in quantity, but also accompanied by multi-source and multi-category characteristics of data. Therefore, traditional data storage and management methods cannot meet current needs. It is imminent to find new technologies that effectively use big data. Hash learning maps data into binary strings through machine learning mechanisms, which can significantly reduce data storage and communication overhead, thereby effectively improving the efficiency of the learn...

AI Technical Summary

Problems solved by technology

[0005] In summary, the existing supervised hashing method is not suitable for large-scale data, and the unsupervised hashing method has no class label information, resulting in low retrieval accuracy

Images