Document Keypoint Extraction via Character Tip Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document alignment systems lack precision in keypoint extraction, affecting the accuracy of image matching in document digitization and comparison processes, particularly when dealing with variations in document templates and filled forms.
Innovation Solution
A processor-implemented method for keypoint extraction that uses Optical Character Recognition (OCR) to determine bounding box coordinates, identifies pre-defined character lists with distinct tips, extracts connected components, and selectively searches for keypoints in the first or last component of each word, ensuring accurate alignment and matching between reference and test images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If centroids of words are used for homography estimation and alignment, then the alignment process can be performed, but the precision is insufficient
Solution Approach 1:
The patent segments the document into multiple words and further into individual characters, extracting centroids from multiple character-level keypoints rather than using a single word centroid. This segmentation approach provides more reference points for homography estimation, improving both alignment precision and matching reliability.
Solution Approach 2:
The patent performs preliminary keypoint extraction and filtering before alignment. It pre-identifies characters with distinct tips (such as 'A', 'V', 'T', 'Y') as reliable keypoints, and pre-computes their centroids. This preliminary preparation ensures that only high-quality keypoints are used in the subsequent alignment process, enhancing precision and reliability.
2Reliability
If punctuation characters are used as keypoints, then keypoint-based alignment can be performed, but the quality of image matching varies based on the number and selection of keypoints
Solution Approach 1:
The patent applies local quality by differentiating between different types of characters and assigning different weights or priorities to them as keypoints. Characters with distinct tips (like 'A', 'V', 'T', 'Y') are identified as having higher local quality for keypoint extraction. This selective approach improves image matching quality by focusing on the most reliable characters while managing selection complexity through clear criteria.
Solution Approach 2:
The patent changes the parameter of keypoint selection from arbitrary or uniform selection to selection based on character shape properties (presence of distinct tips). By changing the selection criterion from general character presence to specific geometric properties, the system achieves more reliable keypoints without significantly increasing complexity, as the geometric property check is computationally efficient.
3Measurement precision
If more keypoints are extracted from documents, then the accuracy of document alignment improves, but the complexity of the keypoint extraction process increases
Solution Approach 1:
The patent applies partial action by not extracting keypoints from all characters, but only from a selected subset of characters with distinct tips. This partial selection approach provides sufficient keypoints for accurate alignment (excessive enough to ensure reliability) while avoiding the complexity of processing every character in the document. The selection criterion of 'distinct tips' provides a clear, efficient filter that balances quantity and quality.
Data Source
AI summary
Keypoint extraction is done for extracting keypoints from images of documents. Based on different keypoint extraction approaches used by existing keypoint extraction mechanisms, number of keypoints extracted and related parameters vary. Disclosed herein is a method and system for keypoint extraction from images of one or more documents. In this method, a reference image and a test image of a document are collected as input. During the keypoint extraction, based on types of characters present in words extracted from the document images, a plurality of words are extracted. Further, all connected components in each of the extracted words are identified. Further, it is decided whether keypoints are to be searched in a first component or in a last component of all the identified connected components, and accordingly searches and extracts at least four of the keypoints from the test image and the corresponding four keypoints from the reference image.