Elastic motion estimation method based on bidirectional iterative optimization and adaptive fusion

Abstract

The invention discloses an elastic motion estimation method based on bidirectional iterative optimization and adaptive fusion. The method comprises the steps: firstly, calculating a translation component of an elastic motion vector through rhombus search, and reducing the sensitivity of elastic motion estimation to a search starting point; secondly, by adopting a forward iterative optimization method based on a Gaussian-Newton method, calculating motion mapping from a reference frame to a current frame, by adopting a reverse iterative optimization method based on a Levenberg-Marquardt method, calculating motion mapping from the current frame to the reference frame, and then providing a bidirectional iterative optimization elastic motion model solving strategy; then, performing linear fusion on motion compensation results of forward and reverse iterative optimization, judging correlation between forward and reverse elastic motion components and object motion by utilizing an inner product of a matching error function and an elastic motion primary function, and performing adaptive fusion on bidirectional elastic motion vectors; and on the basis, performing optimal decision making on the two fusion strategies according to motion compensation errors, so that elastic motion estimation of bidirectional iteration optimization is achieved.

Description

technical field [0001] The invention relates to the field of video coding and compression, in particular to an elastic motion estimation method based on bidirectional iterative optimization and self-adaptive fusion, which has fast operation speed, good motion compensation quality, high convergence efficiency and strong practicability. Background technique [0002] Under the framework of block-based differential prediction + transform closed-loop hybrid coding, motion estimation is an effective time-dimension differential prediction technology, which is widely used in AVS, H.261 / 2 / 3, MPEG-1 / 2 / 4, In video coding standards such as H.264 / AVC, H.265 / HEVC, H.266 / VVC and VC-1, they have played an important role in improving video coding efficiency and improving the subjective and objective quality of video coding. A key technique for the overall rate-distortion performance of an encoder. However, the computing resources consumed by the motion estimation link often account for more...

AI Technical Summary

Problems solved by technology

However, there are still two shortcomings in the motion estimation method based on the elastic model: first, the Gauss-Newton algorithm for solving the elastic motion model needs to calculate partial derivatives, Hessian matrix, inverse matrix, and matrix multiplication in each iteration process , bilinear interpolation and motion compensation errors, whose computational complexity is even higher than the full search of the block translation model, which significantly reduces the practicability of elastic motion estimation in video coding; second, the traditional elastic model-based motion estimation method Most of them use the Gauss-Newton method to solve the motion vector, which is very sensitive to the initial search point, easy to fall into the local optimum, and the convergence efficiency is not high, and it is impossible to obtain satisfactory motion compensation quality and rate-distortion performance

Although this method has achieved good convergence efficiency, its calculation amount is about 65% of the full search based on the block translation model, which is still higher than the computational complexity of representative fast motion estimation methods based on block matching such as TZSearch

[0008] To sum up, there is currently no elastic motion that can achieve an ideal compromise between computational complexity, convergence efficiency, and motion vector accuracy, with fast computing speed, good motion compensation quality, high convergence efficiency, and strong practicability. estimation method

Images