Image Processing Apparatus for High-Resolution OCR on Large Documents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
OCR accuracy is compromised when capturing large-sized paper documents like A4 or A3 due to reduced character resolution and image blurring, necessitating close proximity imaging to improve text recognition.
Innovation Solution
An information processing apparatus that extracts feature amounts from entire and local images, decides the local image position, receives user selections, and converts coordinates to locate and highlight text regions within the document efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If OCR is performed on large-sized paper documents (A4 or A3) captured from a distance, then the entire document can be captured, but OCR accuracy is lowered due to reduced character resolution and image blurring
Solution Approach 1:
The patent divides the document processing into multiple regions: the entire document image is segmented into multiple local images, and text regions are further segmented into individual text elements. This allows OCR to be performed on high-resolution local sections rather than attempting to capture the entire large document at once, thereby maintaining OCR accuracy while covering large areas.
Solution Approach 2:
The patent introduces coordinate system transformation as an additional dimension to solve the contradiction. By establishing coordinate relationships between the entire document image and local images, and between local images and text regions, the system can accurately locate and process text elements at high resolution even in large-sized documents captured from a distance.
2Measurement precision
If images are captured in close proximity to text regions to improve OCR accuracy, then text recognition accuracy improves, but the time required to capture and process the entire document increases
Solution Approach 1:
The patent performs preliminary actions by first capturing the entire document image to identify the document layout and text regions, then pre-processing the image to locate areas containing text. This preliminary analysis allows the system to efficiently direct subsequent close-proximity captures only to relevant text regions rather than capturing the entire document at high resolution, thereby reducing processing time while maintaining OCR accuracy.
Solution Approach 2:
The patent employs dynamic processing where the capture and processing strategy adapts based on the document characteristics. The system dynamically adjusts which regions require close-proximity capture based on the detected text distribution and importance, optimizing the balance between OCR accuracy and processing time for different document types and sizes.
3Productivity
If the entire image is captured at once, then processing speed is improved, but text region identification accuracy decreases due to image size and resolution constraints
Solution Approach 1:
The patent segments the image processing task into multiple stages: first capturing and processing the entire document to identify overall structure, then dividing it into local images for detailed text region identification. This segmentation allows efficient processing of large images while maintaining high accuracy in text region detection by analyzing smaller, higher-resolution sections.
Solution Approach 2:
The patent introduces coordinate transformation systems as intermediaries that link the entire document image with local images and text regions. This intermediary coordinate system allows the system to efficiently process the entire document at low resolution for layout analysis, then accurately locate text regions using coordinate mappings without requiring the entire high-resolution image to be processed at once.
Data Source
AI summary
The information processing apparatus includes an extraction unit configured to extract a feature amount of an entire image and a feature amount of a local image, a decision unit configured to decide a position of the local image in the entire image based on the feature amount of the entire image and the feature amount of the local image, a reception unit configured to receive a selection of an object included in the local image, and a conversion unit configured to convert coordinate of the selected object in the local image into coordinate in the entire image based on the decided position of the local image in the entire image.