Document Deskewing via Text Bounding Box Coordinates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional deskewing techniques for documents with visual aberrations, such as rotations and perspective distortions, are resource-intensive and often fail when document borders are not clearly visible, leading to difficulties in downstream processing tasks like OCR and information extraction.
Innovation Solution
A machine learning model is applied to identify text bounding boxes in a document, determining the coordinates of a document bounding box that encapsulates these aberrations, allowing for the application of transformations to correct them, thereby deskewing the image with reduced computational resources and latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional deskewing techniques are used, then document skew correction is achieved, but computational resources and processing time are excessively consumed
Solution Approach 1:
The patent segments the document processing task into distinct stages: text detection, bounding box coordinate extraction, machine learning model inference for skew angle prediction, and selective deskewing application. This segmentation allows each component to be optimized independently, reducing overall computational overhead while maintaining accuracy.
Solution Approach 2:
The patent replaces traditional mechanical/image-processing-based deskewing methods with a machine learning model that predicts skew angles from text bounding box coordinates. This substitution reduces computational complexity by using statistical patterns rather than intensive image analysis.
2Reliability
If conventional deskewing techniques are used, then skew correction is attempted, but failure occurs when document borders are not clearly visible
Solution Approach 1:
The patent introduces text bounding boxes as an intermediary element that mediates between the image content and the deskewing operation. By using detected text regions as reference points rather than relying on document borders, the system achieves reliable skew correction even when borders are invisible or ambiguous.
Solution Approach 2:
The patent changes the reference parameter from document border coordinates to text bounding box coordinates. This parameter change enables the system to determine skew angles based on the geometric relationships between text regions, which remain detectable even when document borders are not visible.
3Measurement precision
If machine learning model is applied to all text bounding boxes, then accurate deskewing is achieved, but computational latency increases
Solution Approach 1:
The patent applies partial action by using only a subset of detected text bounding boxes for skew angle calculation rather than processing all text regions. This selective approach maintains sufficient accuracy for deskewing while significantly reducing the computational burden and processing latency.
Solution Approach 2:
The patent creates a universal solution where the machine learning model can handle varying numbers of text bounding boxes through standardized coordinate processing. The model is designed to work efficiently whether fed with minimal or maximal text region data, providing consistent performance across different document types.
Data Source
AI summary
A method may include determining, based at least on an image of a document, a plurality of text bounding boxes enclosing lines of text present in the document. A machine learning model may be trained to determine, based at least on the coordinates defining the text bounding boxes, the coordinates of a document bounding box enclosing the text bounding boxes. The document bounding box may encapsulate the visual aberrations that are present in the image of the document. As such, one or more transformations may be determined based on the coordinates of the document bounding box. The image of the document may be deskewed by applying the transformations. One or more downstream tasks may be performed based on the deskewed image of the document. Related methods and articles of manufacture are also disclosed.