Binary Image Area Gradation for Document Orientation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image orientation correction methods, such as OCR, face difficulties in identifying the top and bottom of documents with few characters, like photographs, and struggle to maintain image gradation when using binary images for deep learning-based top-bottom identification, leading to reduced recognition accuracy.
Innovation Solution
The approach involves generating binary images that maintain gradation through preprocessing techniques like error diffusion binarization and scaling using methods like area average or Gaussian filters, allowing for equivalent recognition accuracy to multi-level images within resource-constrained devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If OCR method is used for top-bottom identification, then character-based documents can be processed, but documents with few characters (like photographs) cannot be accurately identified
Solution Approach 1:
The patent changes the input parameters from character-based OCR data to image-based deep learning features. By transforming the document image into a binary image with area gradation and feeding it to a neural network model, the system can process both text-heavy documents and image-heavy documents (like photographs) uniformly, achieving high identification accuracy across different document types.
2Productivity
If binary image is used for deep learning-based top-bottom identification, then processing speed and resource consumption are improved, but image gradation is lost leading to reduced recognition accuracy
Solution Approach 1:
The patent applies local quality by preserving area gradation information in the binary image. Instead of uniform binarization, the error diffusion method is used to maintain local tonal variations as area differences, allowing the neural network to recognize important features while using efficient binary image processing.
Solution Approach 2:
The patent introduces an intermediary processing step between image capture and deep learning analysis. The binary image with area gradation acts as an intermediary representation that retains essential visual information while being optimized for efficient neural network processing, bridging the gap between multi-level images and simple binary images.
3Measurement precision
If multi-level image is used for deep learning, then recognition accuracy is maintained, but resource consumption and processing complexity increase
Solution Approach 1:
The patent uses binary images with area gradation as a simplified, resource-efficient representation that can be quickly processed. While multi-level images provide detailed information, the binary representation with preserved area relationships provides sufficient information for top-bottom identification at much lower computational cost, making it suitable for resource-constrained devices.
Data Source
AI summary
An image forming apparatus includes circuitry. The circuitry generates a binary image having area gradation or a scaled image having area gradation from an image read by a scanner. The circuitry outputs classification of the binary image or the scaled image according to a neural network model learned in advance.