Extended block matching and motion vector estimation algorithm for multi-image motion estimation

Abstract

The invention relates to an extended block matching and motion vector estimation algorithm for multi-image motion estimation. The number p of image areas in segmentation is set, an analysis image block with M * M pixels is defined in the current image frame X, and a search area in the next image frame Y is defined as an L * L block with the position corresponding to the center of each analysis image block as the center; the p motion vectors are determined to minimize the cost function. According to the invention, various fast search technologies are introduced into the extended block matchingalgorithm to reduce the calculation time; a larger motion analysis system is constructed, the system can detect which local motion configuration occurs in a given analysis image block, local motion vector estimation can be combined into comprehensive explanation of scene motion, and the system is applied to actual problems of various motion image processing. Therefore, local minimum points are avoided, and the purposes of quickly searching the optimal motion vector, effectively reducing the number of times of searching the motion vector and reducing the complexity of motion estimation operation are achieved.

Description

technical field [0001] The invention relates to an extended block matching and motion vector estimation algorithm for multi-image motion estimation, and belongs to the technical field of visual matching. Background technique [0002] Although the concept of the existing block matching algorithm is very simple, it works well even under realistic noise conditions, so it is often used in various practical application fields, such as video processing. However, existing patch matching methods do not work well if a given analyzed image patch contains multiple image regions moving in different directions, and / or if the boundaries of a given analyzed image patch do not coincide with object boundaries. [0003] Existing matching methods fall short: [0004] 1) In order to ensure the efficiency and convergence of the algorithm, the search template and search step size can only be changed successively from large to small, that is, rough positioning is performed first, and then gradual...

AI Technical Summary

Problems solved by technology

[0005] 2) The search is blind, and when the step size of the first step is large, it will mislead the search direction;

[0006] 3) Affect the search speed

It cannot be flexibly processed according to the content motion type of the image

That is, no matter what kind of movement it is, start from the origin, search with a large template or a large step size, and then gradually reduce the step size, which is a waste of small movements

Images