Document Segmentation for OCR Using Pixel-Level Text Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical character recognition systems face difficulties in distinguishing text from background sections in noisy images, particularly due to issues like blurriness or scanner quality, which conventional deep learning techniques struggle to address at a pixel-level precision.
Innovation Solution
A computer-implemented method for document segmentation using a trained model that generates bounding boxes around detected text areas within images, masking non-text sections and employing a convolutional neural network with pixel-by-pixel mapping to differentiate between text and background, utilizing synthesized noisy images for training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional deep learning techniques are used for text detection at pixel level, then text detection capability is improved, but performance deteriorates on noisy images with blurriness or scanner quality issues
Solution Approach 1:
The system performs preliminary denoising and enhancement of the input image before feeding it to the text detection model. This preliminary action cleans the image of scanner artifacts, blurriness, and other noise, ensuring the subsequent pixel-level text detection operates on a cleaned version of the image, thereby maintaining high detection precision and reliability even when the original image is noisy.
Solution Approach 2:
The patent introduces an intermediary processing stage between image acquisition and text detection. This intermediary component (the denoising and enhancement module) acts as a mediator that transforms the noisy input image into an enhanced version that is more suitable for accurate text detection, effectively decoupling the detection model from the quality issues of the input image.
2Measurement precision
If pixel-level text differentiation is implemented, then text extraction accuracy is improved, but computational complexity increases
Solution Approach 1:
By performing denoising and enhancement as a preliminary action before pixel-level analysis, the system reduces the computational burden during the actual text detection phase. The enhanced image has fewer artifacts and noise, allowing the pixel-level differentiation to be more efficient and less computationally intensive while maintaining high accuracy.
Solution Approach 2:
The patent segments the image processing into distinct stages: first denoising and enhancement, then pixel-level text differentiation. This segmentation allows each stage to be optimized independently, with the denoising stage preparing the data to reduce computational complexity in the subsequent pixel-level analysis stage.
Data Source
AI summary
An approach to identifying text within an image may be presented. The approach can receive an image. The approach can classify an image on a pixel-by-pixel basis whether the pixel is text. The approach can generate bounding boxes around groups of pixels that are classified as text. The approach can mask sections of an image that where pixels are not classified as text. The approach may be used as a pre-processing technique for optical character recognition in documents, scanned images, or still images.