Document Image Orientation Recognition via Multi-Scale Text Line Filtering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for recognizing the orientation of document images are unsatisfactory when the images are a mixture of text and images, as they fail to achieve high accuracy in processing such mixed-content documents.

Innovation Solution

The method involves generating first and second scale binary images, detecting text lines, calculating similarity between corresponding sections, removing noise text lines, and performing optical character recognition to determine the image orientation based on character orientations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional orientation recognition methods are used on mixed-content document images, then processing can be performed, but recognition accuracy deteriorates due to failure to distinguish text from images

Engineering Contradiction:
Improveorientation recognition accuracyVSAvoidhandling capability for mixed-content documents
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent segments the document image into multiple text line candidates and evaluates each independently using OCR confidence factors. This segmentation allows the system to process text regions separately from image regions, improving orientation recognition accuracy by focusing OCR analysis only on actual text lines rather than treating the entire mixed-content image uniformly.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality assessment by evaluating OCR confidence factors for each detected text line candidate individually. Regions with high OCR confidence are identified as genuine text, while regions with low confidence are treated as images or noise. This local differentiation enables accurate orientation recognition in mixed-content documents by applying text-specific analysis only where text is present.

Inventive Principle:
Principle #3Local quality

2Measurement precision

If multiple text line candidates are detected and evaluated using OCR for each candidate, then orientation recognition accuracy is improved, but processing time increases

Engineering Contradiction:
Improveorientation recognition accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent implements partial action by performing OCR evaluation on multiple text line candidates but using confidence factor thresholds to prune the search space. Instead of exhaustively processing all possible orientations for all candidates, the system performs OCR on candidates and uses confidence scores to eliminate unlikely options, thereby reducing processing time while maintaining high accuracy through selective evaluation.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The patent employs feedback mechanisms where OCR confidence factors from initial evaluations are used to guide subsequent processing decisions. High-confidence text line detections reinforce the current orientation hypothesis, while low-confidence detections trigger re-evaluation or elimination. This feedback loop enables accurate orientation determination without requiring exhaustive processing of all candidates.

Inventive Principle:
Principle #23Feedback

3Measurement precision

If text line detection is performed on full-resolution images, then detection accuracy is improved, but processing speed decreases

Engineering Contradiction:
Improvetext line detection accuracyVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent segments the full-resolution document image into multiple text line candidate regions before performing detailed OCR analysis. This preliminary segmentation at reduced resolution or through rapid detection algorithms identifies potential text regions, allowing subsequent high-accuracy OCR processing to be applied only to these segmented candidates rather than the entire image, thereby maintaining detection accuracy while improving processing speed.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP2660753B1Image processing method and apparatus
Publication Date: 2019.10.16 FUJITSU LTD
  • EP2660753B1 patent drawingFigure 1
  • EP2660753B1 patent drawingFigure 2a~2b
  • EP2660753B1 patent drawingFigure 3a~3c

AI summary

An image processing method and apparatus is provided. The image processing method includes steps of: generating a first scale binary image from an image, wherein the first scale is smaller than the original scale of the image; detecting at least one text line in the image based on the first scale binary image; generating a second scale binary image from the image, wherein the second scale is larger than the first scale; for each text line, calculating a similarity between corresponding sections in the first scale binary image and the second scale binary image, and removing the text line for which the similarity is lower than a predetermined level; for one or more of the remaining text line(s), performing OCR on corresponding section(s) in the second scale binary image to determine character orientation(s) of corresponding text line(s); and determining the orientation of the image according to the determined character orientation(s).