Feature-Based Image Alignment with Similarity-Guided Offsets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image alignment methods based on optical flow are inefficient due to high computational costs, and deformable convolutional layers lack initial values, leading to unstable alignment results, especially with poor image quality.
Innovation Solution
An image alignment method using feature-based alignment, incorporating a correlation layer to predict offsets with similarity features as guidance, and employing deformable convolutional layers to stabilize the alignment process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If optical flow-based image alignment is used, then alignment can be achieved, but computational cost is high and efficiency is low
Solution Approach 1:
The patent extracts and utilizes only the correlation information between feature points rather than computing the complete optical flow field. By taking out the essential similarity features and related feature points, the method achieves alignment without the heavy computational burden of full optical flow calculation, thus improving efficiency while reducing energy consumption.
Solution Approach 2:
The patent segments the image alignment task into discrete feature point matching operations. Instead of processing the entire image as a continuous field, it divides the problem into individual feature point correlations, allowing parallel processing and significantly reducing computational complexity while maintaining alignment accuracy.
2Adaptability or versatility
If deformable convolutional layers are used without initial values, then alignment flexibility is improved, but alignment stability deteriorates especially with poor image quality
Solution Approach 1:
The patent performs preliminary action by pre-calculating and providing initial offset values before the deformable convolutional layer processes the image. These initial values serve as a stable foundation that guides the subsequent deformation process, ensuring alignment stability even when image quality is poor, while still allowing the deformable layer to adapt to different scenarios.
Solution Approach 2:
The patent introduces similarity features as an intermediary between the input image and the deformable convolutional layer. These similarity features act as a mediator that provides reliable initial offset information, stabilizing the alignment process before the flexible deformation occurs, thus maintaining reliability while preserving adaptability.
Data Source
AI summary
An image alignment method and device, relating to the technical field of image processing. The method comprises: obtaining a target feature comprising feature points corresponding to pixel points in a target image and a reference feature comprising feature points corresponding to pixel points in a reference image; obtaining a similarity feature according to the target feature and the reference feature, the similarity feature comprising the similarity between the feature points in the target feature and a corresponding related feature point; predicting a convolutional layer according to the similarity feature, the target feature, and an offset, and obtaining an offset of the target feature and the reference feature; and aligning the reference feature with the target feature according to the offset and the deformable convolutional layer.