Face capturing and recognition method based on face quality evaluation

Abstract

The invention discloses a face capture recognition method based on face quality evaluation. The invention adopts depth learning to carry out face quality evaluation score regression training on threetypical data sets with quality grade difference, acquisition of Face Quality Evaluation Model, the model is applied to face capture system, In order to reduce the system load without losing accuracy,the real-time received video stream is divided into several capture cycles according to the fixed number of frames, and the capture quality conditions are dynamically adjusted in each capture cycle, and the highest quality face in a period of time is output for face recognition algorithm to recognize. The invention adopts the technology of depth learning to score face quality, cooperates with facecapture to carry out low-quality face filtering and optimal face selection, thereby better serving the face recognition system, and realizes lower system hardware resource occupancy and higher face recognition rate at the same time.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and relates to a human face snapshot recognition method based on human face quality evaluation. Background technique [0002] The input of face recognition in the video surveillance scene is a large number of captured face images output by the intelligent face capture system. Due to the uncontrollability of the scene, the quality of multiple pictures of the same person in a non-ideal environment is uneven. Common influencing factors are as follows: Too low brightness contrast, too biased facial posture, facial occlusion, drastic expression changes, low resolution, image blur and noise, etc. In order to improve the recognition accuracy, reduce the false recognition rate and improve the system performance, it is necessary to use the face quality evaluation for selection. [0003] Chinese invention patent: CN20111017191.X "Real-time monitoring method for face image quality of customer acqui...

AI Technical Summary

Problems solved by technology

However, in these methods, the acquisition of specific image quality label values ​​required for training face quality evaluation still needs to be given by the above-mentioned traditional methods, or a large amount of labor costs are invested in labeling, which results in a more complicated process, or longer Consumption of time and more manpower

However, in CN201711180270.5 "Face Image Quality Evaluation Method and Device Based on Face Comparison", the similarity of face comparison between face images and standard reference images is used as the quality reference tag value. This method is directly oriented to recognition. It avoids the above defects of label acquisition, but at the same time brings new problems: firstly, for the large amount of data required for deep learning training, it is difficult to obtain and select a large number of reference pictures; another consideration is the quality evaluation of the training The model may only serve a specific face recognition comparison model better, but the effect of the optimal quality face selected for different recognition models will be worse, so it is not universal

Images