Image Alignment Using Block Partitioning and Iterative NCC
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image alignment methods face challenges in accurately aligning photos with templates, especially when features are ambiguous or when documents are not flat, leading to difficulties in extracting information from filled-in forms captured by mobile devices.
Innovation Solution
The method involves partitioning the template into blocks, using an image processor to match features, determine displacement vectors, and apply normalized cross-correlation to select the best displacement vector, iteratively combining blocks to remove ambiguity and align the image with the template.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If feature matching is performed on the entire template at once, then alignment speed is improved, but alignment accuracy deteriorates due to ambiguous features
Solution Approach 1:
The template is divided into multiple blocks, and feature matching is performed independently on each block rather than on the entire template at once. This segmentation allows the system to handle ambiguous features more effectively by localizing matches to smaller regions, thereby maintaining both speed and accuracy.
2Loss of information
If all template features are used for alignment, then completeness of information is improved, but reliability deteriorates due to ambiguous and extra features
Solution Approach 1:
The system extracts and identifies ambiguous features and extra features separately from the matching process. By detecting features that do not have clear correspondences or are present only in one image, the system can exclude these unreliable features from the final alignment calculation, thereby improving reliability while maintaining information completeness.
Solution Approach 2:
The system employs an iterative process where alignment results are used to refine feature matching. Through multiple passes, the system progressively improves alignment accuracy by using feedback from previous iterations to better distinguish between valid and ambiguous features.
3Adaptability or versatility
If feature matching is performed on non-flat documents, then adaptability is improved, but measurement precision deteriorates due to perspective distortion
Solution Approach 1:
The system transitions from 2D feature matching to 3D spatial reasoning by incorporating depth information and perspective transformations. This allows the system to account for the curved or folded nature of non-flat documents, maintaining measurement precision even when documents are not perfectly flat.
Data Source
AI summary
According to exemplary systems and methods, a template is partitioned into blocks using an image processor. An image is scanned using an optical scanner. The image is aligned to the template. Image features are matched with template features in the blocks of the template. Displacement vectors are identified for differences of the image features from the template features. Normalized cross correlation (NCC) is determined between blocks of the image and each block of the template using the image processor. Peaks in the NCC are identified. A displacement vector is selected for a peak with highest NCC for each the block of the template. Ambiguous template features are identified in the image based on the displacement vector. Blocks are iteratively combined and NCC determined to remove the ambiguous template features.