Adaptive motion search method based on rotational scaling

Abstract

The invention discloses an adaptive motion search method based on rotational scaling. The method comprising the steps of selecting a reference emotion parameter from candidate reference blocks, and initializing search; setting an initial search step size and starting search; searching a translation motion parameter, to acquire an optimal point A; using A as an initial point, to search rotation andscaling parameters, and acquiring a current optimal point B; meanwhile, performing gradient descent type search on the four parameters until the an error is reduced to be less than a preset thresholdor the number of iterations is more than a set upper limit; separately acquiring absolution values of the four parameters with the last iteration step size as the step size, separately performing two-point type search in positive and reverse directions, and thus acquiring a final optimal value. According to the method provided by the invention, the motion of an object can be better depicted, andthe more proper reference block can be found; the motion details of the object can be more accurately depicted, and the search loop can be more quickly skipped out when moving away from the search area; the problems of insufficient accuracy and pixel value fluctuation can be avoided, and thus the pixel value is more accurate when in non-integral point.

Description

technical field [0001] The invention belongs to the field of video coding and scene detection, and relates to a motion search method based on rigid rotation and scaling of objects in video coding, in particular to an adaptive motion search method based on rotation and scaling, which can not only be used to reduce coding codewords, It can also be used as an auxiliary technology for scene detection and reconstruction, and then become unmanned driving, map collection, landform surveying and mapping, and three-dimensional scene reconstruction. . Background technique [0002] In the traditional search process of the reference block between video frames, the search is mainly aimed at 2D translation. It is based on the assumption that all motions of objects can be approximated as rigid body translational motions on the viewing plane. However, this assumption is relatively rough. There is no absolute rigid body in the real world, and the movement of objects is often more complicat...

AI Technical Summary

Problems solved by technology

How to encode more efficiently in higher-definition video poses many challenges, two of which are of great research value: one is the motion description of objects, as mentioned above, in traditional motion search, objects are approximately described as A rigid body that translates on a two-dimensional plane. However, the difference in this approximate effect in high-definition videos will be more exposed than in low-bit-rate videos. Therefore, a more suitable motion model is required to be more accurate. On the other hand, in a high-bit-rate video screen, the residual quantization usually uses a smaller step size, which means that a large and irregular residual description will require a very high code Therefore, how to express the predicted value more accurately is also an important means to improve the coding efficiency

Images