A face attribute association method based on person re-identification

Abstract

The invention discloses a face attribute association method based on re-identification of pedestrians, aiming at the problem of masked and disguised identity recognition. First, a global feature learned together with local features is extracted, the shortest path between two sets of local features is calculated to perform matching, and several pedestrian images most similar to the target person are found. Then, the face position in the corresponding pedestrian image is located, and a face image quality assessment method based on fragmentation is used to quantify the face image quality into the face similarity probability, so as to select the most recognizable face image. Finally, a trusted identity recognition metric model is built, and the face image with the highest confidence is used as the surrogate face of the masked target. The invention converts the masked face recognition problem into the correlation and correspondence problem between the masked face and the displayed human face of the same object, and provides an effective means for the physical identification of the masked and camouflaged target.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and relates to a face attribute association method in video data, in particular to a face attribute association method based on pedestrian re-identification. technical background [0002] Face is the most important basis for identification. Google's FaceNet, the most advanced face recognition technology, has an accuracy rate of 99.63% on the LFW standard data set. However, for the motive of deliberately hiding their identities, criminals often wear sunglasses, masks, hats, turbans and other accessories to cover their faces at the crime scene, or use hoods or even 3D masks to disguise their faces. Behavior poses a huge challenge for identification. [0003] Compared with the influence of factors such as expressions and postures, it is more difficult to deal with facial occlusion factors such as headgear, masks, sunglasses, masks, and hats. In 2016, scholars from Carnegie Mellon Universit...

AI Technical Summary

Problems solved by technology

[0004]Although the current occluded face recognition has made positive progress, it is only limited to a single mild occlusion situation such as glasses, masks, and false beards. For wearing sunglasses at the same time For heavy occlusions such as masks, the recognition rate will drop sharply, and it will be completely invalid for masked face recognition

Images