OCR Bounding Box Clustering for Text Stream Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional optical character recognition (OCR) systems produce scattered text blocks that are difficult for computers to assemble into a coherent text stream, degrading computing device capability and performance, as they require human intervention to align text boxes correctly.
Innovation Solution
A method involving three stages: creating a weight matrix from real-world characteristics of OCR bounding boxes to determine their likelihood of belonging to the same cluster, clustering using methods that don't require the number of clusters as input, and ordering clusters and boxes within clusters based on y-coordinate and x-coordinate to produce a proper text stream.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OCR systems are used to recognize text, then text recognition is achieved, but the output consists of scattered text blocks that are difficult to assemble into a coherent text stream
Solution Approach 1:
The patent introduces an intermediary processing system that receives scattered OCR bounding boxes and automatically clusters them into coherent text streams. This intermediary system uses weight matrices and clustering algorithms to bridge the gap between scattered OCR output and organized text, eliminating the need for manual intervention while improving alignment quality.
Solution Approach 2:
The system performs self-service by automatically assembling scattered text blocks into coherent streams without human intervention. The clustering algorithm autonomously determines the correct ordering of text blocks based on spatial relationships and content coherence, making the system self-sufficient in resolving the assembly difficulty.
2Measurement precision
If manual intervention is used to align text boxes correctly, then text stream quality improves, but computing device capability and performance are degraded
Solution Approach 1:
The patent replaces the mechanical system of manual text alignment with an automated computational system. Instead of human operators manually ordering text blocks, the system uses weight matrices, clustering algorithms, and coherence metrics to automatically assemble text streams, thereby eliminating the performance degradation associated with manual intervention while maintaining high alignment quality.
3Extent of automation
If clustering methods requiring number of clusters as input are used, then clustering can be performed, but the system cannot automatically determine the optimal number of clusters
Solution Approach 1:
The patent changes the parameter requirements of the clustering method by transitioning from approaches that require the number of clusters as input to methods that automatically determine this parameter. The system uses weight matrices and coherence-based metrics to dynamically identify the optimal number of clusters, transforming a fixed parameter requirement into a dynamically determined value.
4Productivity
If scattered OCR blocks are produced, then OCR recognition is completed, but the text blocks are hard to consume by algorithms and systems
Solution Approach 1:
The patent applies preliminary action by organizing scattered OCR blocks into coherent text streams immediately after recognition, before the data is consumed by downstream algorithms. This pre-processing step transforms the scattered output into a structured format that is easily consumable by subsequent processing systems, eliminating the need for manual reorganization and improving overall system efficiency.
Data Source
AI summary
Representative embodiments disclose mechanisms to create a text stream from raw OCR outputs. The raw OCR output comprises a plurality of bounding boxes, each bounding box defining a region containing text which has been recognized by the OCR system. A weight matrix is calculated that comprises a weight for each pair of bounding boxes. The weight representing the probability that a pair of bounding boxes belongs to the same cluster. The bounding boxes are then clustered along the weights. The resulting clusters are first ordered using an ordering criteria. The bounding boxes within each cluster are then ordered according to a second ordering criteria. The ordered clusters and bounding boxes are then arranged into a text stream.