Image Alignment Using Displacement-Based Feature Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing techniques fail to accurately align images with depth variations, leading to artifacts in combined images when objects with different depths are present, as they do not account for regional misalignments effectively.
Innovation Solution
An image processing device and method that calculates the degree of alignment using projection conversion matrices, classifies feature regions with abnormal displacements, and applies filter processing to determine abnormal regions, allowing for the generation of accurate combined images by using evaluation scores based on displacement differences between adjacent feature regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single projection conversion matrix is used to align images, then the alignment process is simple and fast, but alignment accuracy deteriorates when images contain objects with different depths
Solution Approach 1:
The image is divided into multiple regions corresponding to different depth planes. Instead of applying a single projection conversion matrix to the entire image, the patent calculates separate projection conversion matrices for each region based on its depth characteristics. This segmentation allows each region to be aligned with its own optimal transformation, resolving the contradiction between simple alignment processes and accurate multi-depth alignment.
2Manufacturing precision
If multiple projection conversion matrices are calculated for different planes, then alignment accuracy for images with depth variations is improved, but the processing complexity and computational load increase
Solution Approach 1:
Different projection conversion matrices are applied to different regions of the image according to their local depth characteristics. Each region receives a transformation optimized for its specific plane, rather than applying a uniform transformation globally. This local quality approach improves alignment accuracy for multi-depth scenes while managing complexity by focusing computational effort where depth variations exist.
Solution Approach 2:
The patent performs preliminary classification of feature points into different depth planes before calculating projection conversion matrices. By pre-organizing feature points according to their depth information, the system prepares the data structure needed for efficient multi-matrix calculation, reducing the overall processing complexity despite using multiple matrices.
3Ease of manufacture
If feature points from all regions are used to calculate projection conversion matrices, then the calculation is straightforward, but abnormal feature points from different planes cause misalignment artifacts
Solution Approach 1:
The patent extracts and separates feature points belonging to different depth planes before calculating projection conversion matrices. By taking out feature points from specific planes and processing them independently, the system avoids the contamination of abnormal feature points from other planes, thereby improving alignment reliability while maintaining reasonable calculation simplicity through structured processing.
Solution Approach 2:
The system performs preliminary classification and filtering of feature points to identify and separate abnormal feature points before they can corrupt the projection conversion matrix calculation. This preliminary action ensures that only appropriate feature points from the correct plane are used in each matrix calculation, improving reliability without significantly complicating the overall process.
Data Source
AI summary
An image processing device includes circuitry configured to: calculate a displacement of each of a plurality of corresponding feature regions between a reference image and a base image; calculate, as a evaluation score, difference value between displacements of two feature regions adjacent to each other in at least one of the up/down direction, the left/right direction, and the oblique 45° direction; determine an abnormal region on the basis of the score; classify other feature regions excluding the abnormal feature region; calculate a projection conversion matrix by using the displacement of the other feature regions and the result of the classification; calculate a degree of alignment of each pixel of the reference image with respect to each pixel of the base image by using the matrix; and generate a combined image by combining the reference image converted based on the degree of alignment with the base image.