Face image anti-shake method, device and equipment and readable storage medium

Abstract

The invention provides a face image anti-shake method and device, equipment and a readable storage medium. The method comprises the following steps: acquiring a plurality of face feature points in a current image frame; determining a displacement vector of the face feature point based on the current image frame and at least two adjacent image frames of the current image frame; determining a clustering space corresponding to the human face feature points, and clustering the human face feature points in the clustering space according to the displacement vector, wherein the clustering space comprises the human face feature points and human face feature points in a specified area around the human face feature points; and performing anti-shake processing on the face feature points according to the clustering result.

Description

technical field [0001] The present application relates to the technical field of image processing, in particular to a face image stabilization method, device, equipment and readable storage medium. Background technique [0002] With the development of technology, deep learning technology is used in all aspects of life / industry by training models obtained from big data models. In the face application scenario, the face itself has structural and stable characteristics, and the application is very successful. [0003] Whether it is a deep learning algorithm or a traditional face feature point algorithm, the pixels in the output result will inevitably jitter. Jitter is usually caused by external disturbances such as random noise, the background of the image, or changes in lighting when multiple images are taken. In the traditional face anti-shake technology, different filtering algorithms are usually used for filtering processing, such as Kalman filtering, median filtering, Ga...

AI Technical Summary

Problems solved by technology

In the traditional face anti-shake technology, different filtering algorithms are usually used for filtering processing, such as Kalman filtering, median filtering, Gaussian filtering and other algorithms, but these filtering algorithms usually do not pay attention to the output results that are originally in the face feature points Whether it is normal movement or jitter, the output result is treated as jitter for filtering processing. If the weight of the de-jitter algorithm is large during processing, the processing time will be longer and the image will be delayed. If the weight is small, it will bring inevitable jitter.

Theoretically, when the face is still, the feature points output by multiple images processed by the filtering algorithm should have no obvious jitter in the visual effect, and when the face moves, it is necessary to maintain the anti-shake effect while not being too large. delay, but the current general-purpose filtering algorithm cannot achieve

Images