Block-Based Image Registration with Depth-Aware Motion Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image registration methods, such as block matching and feature point-based matching, fail to accurately account for local moving objects, while optical flow methods often blur edges, leading to reduced registration accuracy.
Innovation Solution
An image registration method that divides images into blocks, determines similar blocks using a similarity algorithm, calculates movement amounts, performs clustering based on depth values and position information, and moves blocks to generate a registered image when predetermined conditions are met.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If block matching or feature point-based matching is used, then the method is simple to implement, but it can only obtain a global motion model without considering local moving objects, reducing registration accuracy
Solution Approach 1:
The patent segments the image into multiple blocks and performs matching on each block independently. This allows the system to capture both global motion trends and local movements of specific objects, thereby improving registration accuracy while maintaining the simplicity of block-based methods.
Solution Approach 2:
The patent applies different processing approaches to different regions of the image based on their characteristics. By identifying moving objects and applying local motion compensation specifically to those regions while using global motion models for stable regions, the system achieves high accuracy without excessive complexity.
2Adaptability or versatility
If optical flow method is used, then local moving objects can be captured, but edge information is processed with insufficient accuracy causing blurred edges, reducing registration accuracy
Solution Approach 1:
The patent divides the image into blocks and processes each block separately using optical flow techniques. This segmentation allows edge preservation within each block while capturing local movements, avoiding the edge blurring problem that occurs when optical flow is applied to the entire image at once.
Solution Approach 2:
The patent introduces block-based motion estimation as an intermediary step before applying optical flow. This intermediary approach provides a coarse motion field that guides the optical flow calculation, preventing edge blurring while still capturing local movements accurately.
3Device complexity
If block matching is used to obtain movement amounts, then computational complexity is low, but sparse movement amounts are obtained which reduces registration accuracy
Solution Approach 1:
The patent combines block matching and optical flow methods in a hybrid approach. Block matching provides a coarse motion field with low computational complexity, while optical flow refines this field to provide dense movement amounts. This merging of methods achieves both low complexity and high accuracy.
Solution Approach 2:
The patent applies optical flow only to regions where block matching indicates significant motion, rather than computing optical flow for the entire image. This partial application reduces computational complexity while still obtaining dense movement amounts in the critical regions, improving registration accuracy without excessive computation.
Data Source
AI summary
An image registration method includes: obtaining a first image and a second image; dividing the first image into blocks to obtain image blocks of the first image; determining, based on a pixel value of each image block, a similar image block of the image block from the second image by invoking a predetermined image block similarity algorithm; calculating, based on position information of the image block and position information of the similar image block, a movement amount of the image block of the first image; performing a clustering on the first image based on a depth value of the first image, the movement amounts of the image blocks, and position information of the first image, to generate clustered image blocks; and moving, when the clustered image blocks satisfy a predetermined condition, the image blocks based on the movement amounts of the image blocks, to obtain a registered image.