Footprint pressure image retrieval method

Abstract

The invention discloses a footprint pressure image retrieval method. The method comprises the following steps: 1, collecting and preprocessing a footprint pressure image data set; 2, establishing a feature extraction network composed of K layers of multi-scale self-attention convolution modules; 3, establishing an incomplete scoring module composed of a global feature branch and an incomplete scoring mask branch; step 4, establishing a feature comparison module composed of a common visible feature extraction module, a local feature pooling module and a triple loss function; and step 5, carrying out network training, parameter optimization and testing. The multi-scale self-attention convolution footprint pressure image retrieval method is adopted, discriminative features of footprint pressure image retrieval can be effectively extracted, and meanwhile, for the problem that footprint images are incomplete, an incomplete scoring model is adopted, so that the attention of a network to incomplete parts can be reduced, and the influence of the incomplete image on the retrieval process is effectively inhibited.

Description

technical field [0001] The invention relates to the fields of image retrieval, image processing and deep learning, and is a footprint pressure image retrieval method based on multi-scale self-attention convolution. Background technique [0002] In the field of biometrics, various external and internal features of people such as face, iris, fingerprints, gait, etc. can be used for human identification. At present, in the process of evidence extraction at a crime scene, on-scene footprints are one of the most common important criminal clues, and they play an irreplaceable role in case analysis. [0003] Studies have shown that because each person's footprint has its own characteristics, it is unique. In the previous footprint comparison work, most of the fingerprint experts manually compared the collected criminal suspect's footprints with those at the crime scene for personnel verification. However, due to the complexity of footprint image comparison and the The incompleten...

AI Technical Summary

Problems solved by technology

In the previous footprint comparison work, most of the fingerprint experts manually compared the collected criminal suspect's footprints with those at the crime scene for personnel verification. However, due to the complexity of footprint image comparison and the The incompleteness of the data often exists, which greatly increases the difficulty of comparison

In addition, footprint comparison mostly relies on the knowledge and experience of experts, which often has a large subjectivity, and there is still a certain gap between the automatic comparison and recognition of footprints.

Images