Document Image Template Matching via Graph Structure Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Doc-Image Template Matching techniques face challenges such as distorted document regions, variations in text content, unexpected graphical elements, and varying table or text block lengths, which affect the accuracy of automated document processing by limiting the analysis to visual information only and failing to utilize structural and textual elements effectively.
Innovation Solution
The method integrates visual and structural information with textual information to improve template matching by generating knowledge graphs that represent visual and structural elements along with textual content, allowing for more accurate field template matching and document classification, even in cases with shifted block structures or diverse document types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If image hashing is used to compare documents to base templates, then the process is simple and fast, but the accuracy is limited due to visual information only
Solution Approach 1:
The patent segments the template matching process into multiple stages: initial filtering using image hashing for speed, then detailed comparison using structural elements and concepts for accuracy. This multi-level segmentation allows the system to quickly eliminate non-matching templates while performing thorough analysis only on promising candidates.
Solution Approach 2:
The patent transitions from two-dimensional visual pixel comparison to a multi-dimensional analysis that includes structural elements, textual content, and conceptual relationships. By adding these additional dimensions of analysis, the system achieves higher accuracy without completely sacrificing processing efficiency.
2Device complexity
If only visual information is used for template matching, then the process is straightforward, but it fails to handle distorted regions, text variations, and structural differences
Solution Approach 1:
The patent divides the document into distinct structural elements (headers, body text, tables, images) and analyzes each separately. This segmentation allows the system to handle distortions and variations in different regions independently, improving reliability without requiring complex global processing.
Solution Approach 2:
The patent introduces structural elements and conceptual representations as intermediaries between the raw visual image and the template matching process. These intermediaries capture the essential semantic and structural information while being robust to visual distortions, text variations, and layout differences.
3Measurement precision
If structural elements and textual content are analyzed in detail, then template matching accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary analysis by extracting structural elements and generating concept representations before the actual template matching. This preliminary processing organizes the information in a way that accelerates the subsequent matching process, reducing the computational burden during the critical comparison phase.
Solution Approach 2:
The patent applies partial detailed analysis only to the most promising template candidates identified by initial filtering. Rather than performing exhaustive detailed analysis on all templates, the system focuses computational resources on a subset, achieving high accuracy while minimizing processing time.
Data Source
AI summary
Computer implemented methods, systems, and computer program products include program code executing on a processor(s) that merges a document comprising multiple pages into a single document image. The program code processes the single document image to identify structural elements and textual content. The program code compares the structural elements of the single document image to other structural elements of a group of document templates stored in a database to identify a subset of the group of documents templates with a threshold number of similarities to the single document image. The program code generates, from the single document image, a graph structure representing the document, where the graph structure comprises visual information and connections related to the structural elements and concepts comprising the textual content. The program code uses the structure to identify a document template that is a closest match to the document.