Face sequence optimal frame extraction method and device and storage medium

Abstract

The embodiment of the invention discloses a face sequence optimal frame extraction method and device and a storage medium. The method comprises the following steps: acquiring video data, and making a face quality evaluation data set according to the video data; dividing the face quality evaluation data set into a training set and a test set; training a selected polynomial regression algorithm by adopting the training set to obtain an optimal frame extraction model; and performing quality judgment on the test set by adopting the optimal frame extraction model to obtain an optimal frame picture of the face sequence. According to the embodiment of the invention, the face is not filtered by an absolute threshold value, the polynomial regression algorithm focuses on the extraction of the relatively optimal face in the face sequence, and one optimal face is certainly screened from each face sequence, so that the missing detection of the face is avoided, and the computing power of the system is small.

Description

technical field [0001] The invention relates to the technical field of computer software, in particular to a method, device and storage medium for extracting optimal frames of a human face sequence. Background technique [0002] In practical applications, face recognition systems usually need to obtain the best-quality face pictures in the face sequence for subsequent recognition and feature extraction. The quality of the optimal face selection directly affects the quality of the facial features extracted by the face recognition system, and will also affect the subsequent related downstream tasks such as face clustering, recognition and matching. Therefore, an efficient face quality evaluation algorithm has positive significance for the entire face system. [0003] At present, the usual face screening logic is as follows: (1) The optimal face screening in the face sequence is performed by ranking the face detection confidence; however, there is no strict positive correlatio...

AI Technical Summary

Problems solved by technology

[0004] Furthermore, using a single-dimensional feature cascade to set a threshold to filter faces cannot comprehensively judge the quality of faces, and it is impossible to screen out the best quality faces when a certain feature of the face sequence does not meet the standard (or all meet the standard).

For example, if a feature condition of all faces in a face sequence is less than the set threshold, all faces in the sequence will be judged as invalid faces, which may easily cause missed detection of faces.

Or if all feature thresholds meet the set threshold requirements, the faces in the sequence will be judged as valid faces, which will cause repeated calculations and increase the computing power burden of the system

Images