Extremely low bit rate face video mixed compression method and system based on bidirectional frame prediction

Abstract

The invention discloses a bidirectional frame prediction method for a face video. The method comprises the following steps: sampling three frames of images of the face video in a time domain; obtaining key point thermodynamic diagrams corresponding to the three frames of images respectively; and in combination with the three frames of images and the key point thermodynamic diagram, reconstructing a middle frame of image. According to the invention, synthesis of a smooth and fluent video in a time domain is facilitated, and subjective experience of viewers is improved; the invention further discloses an extremely-low-bit-rate face video mixed compression method based on the method, and the advantage that a traditional coding technology can carry out pixel-level accurate recovery is combined with the advantage that a deep learning technology can carry out high-quality reconstruction based on brief information. The purpose of synthesizing a high-quality face video at an extremely low bit rate is achieved.

Description

technical field [0001] The invention relates to the field of face video compression, in particular to a hybrid compression method and system for extremely low bit rate face video based on bidirectional frame prediction. Background technique [0002] Video calls are a common way for people to communicate daily, and the usage scenarios of video calls are increasing day by day. In the case of poor network quality, video calls are often forced to be interrupted, which seriously affects the user's calling experience. Therefore, how to achieve stable ultra-low bit rate video calls in the case of poor network quality is a research hotspot, and the research on face video compression technology has become increasingly important. Existing face video compression methods can be divided into two categories: traditional encoding methods and deep learning-based methods. [0003] Traditional video coding methods have the advantage of pixel-level accurate restoration, with VVC being the st...

AI Technical Summary

Problems solved by technology

Some of these methods use static reference frames, which makes it difficult to guarantee good performance in the case of scene switching or large pose changes, and cannot achieve high-fidelity restoration; some add synthetic frames to the reference frame pool, which is easy to cause error accumulation and cannot Perform pixel-level accurate restoration, which happens to be the strength of traditional encoding methods

In addition, most of the existing face video compression methods based on deep learning are difficult to achieve dynamic adjustment of bit rate, which limits their practicability

Images