Text Extraction from Noisy Images Using Multi-Filter Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional OCR tools are ineffective in extracting textual content from images with noise, such as handwritten characters, smudge marks, and faded text, often producing erroneous results and requiring special training with noisy image samples.
Innovation Solution
A method and system that preprocesses image data to create readable copies, applies multiple noise filters in parallel or series to remove noise, performs image data recognition using OCR tools, ranks words based on predefined parameters, and selects the highest ranked words to produce output text, without the need for special training on noisy images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional OCR tools are used to extract text from images with noise, then the extraction process is simple, but the accuracy of text extraction deteriorates significantly
Solution Approach 1:
The patent segments the text extraction process into distinct stages: preprocessing the image data, applying multiple noise filters to remove different types of noise, performing OCR on the cleaned images, and then post-processing the extracted text. This segmentation allows each stage to be optimized independently, improving overall accuracy while maintaining operational simplicity through automated multi-stage processing.
Solution Approach 2:
The patent introduces an intermediary cleaning stage between the original noisy image and the OCR process. Multiple noise filters act as intermediaries that progressively remove different types of noise (handwritten characters, smudge marks, faded text) before the text extraction occurs, thereby improving accuracy without requiring the OCR tool itself to be modified.
2Measurement precision
If conventional OCR tools are trained with noisy image samples to improve accuracy, then the text extraction accuracy improves, but the complexity of the system increases
Solution Approach 1:
The patent applies preliminary noise removal actions before the OCR process occurs. By preprocessing the images to remove handwritten characters, smudge marks, and faded text through multiple filtering stages, the system prepares clean input for standard OCR tools, achieving high accuracy without modifying or retraining the OCR system itself.
Solution Approach 2:
The patent extracts and removes the noise components (handwritten characters, smudge marks, faded text) from the images before text extraction. This separation of noise removal from text extraction allows the use of standard OCR tools while achieving accuracy comparable to specialized trained systems.
3Measurement precision
If multiple noise filters are applied to remove different types of noise, then the text extraction accuracy improves, but the processing time increases
Solution Approach 1:
The patent segments the noise removal process into multiple specialized filters that target different noise types (handwritten characters, smudge marks, faded text). Each filter is optimized for its specific noise type, allowing efficient parallel or sequential processing that removes noise effectively without excessive processing time.
Solution Approach 2:
The patent adjusts processing parameters dynamically based on the detected noise types in the image. By identifying which noise filters are needed based on the specific image characteristics, the system avoids applying unnecessary filters, thereby reducing processing time while maintaining high accuracy.
Data Source
AI summary
A method for extracting text from an image data is disclosed. The method includes pre-processing, via a processor, the image data to obtain a readable image data. The method further includes filtering, via the processor, a plurality of copies of the readable image data using a plurality of noise filters to obtain a corresponding plurality of noise removed images. Yet further, the method includes performing, via the processor, image data recognition on each of the plurality of noise removed images to obtain a text copy associated with each of the plurality of noise removed images. Moreover, the method includes ranking, via the processor, each word in the text copy associated with each of the plurality of noise removed images based on a predefined set of parameters. Finally, the method includes selecting, via the processor, highest ranked words within the text copy associated with each of the plurality of noise removed images to obtain output text for the image data.